Computational Intelligence Theories Applications

and Future Directions Volume I Nishchal K. Verma

pdf download
https://textbookfull.com/product/computational-intelligence-theories-applications-and-future-
directions-volume-i-nishchal-k-verma/

★★★★★ 4.8/5.0 (42 reviews) ✓ 87 downloads ■ TOP RATED

"Amazing book, clear text and perfect formatting!" - John R.

DOWNLOAD EBOOK
 Computational Intelligence Theories Applications and Future
Directions Volume I Nishchal K. Verma pdf download

TEXTBOOK EBOOK TEXTBOOK FULL

Available Formats

■ PDF eBook Study Guide TextBook

EXCLUSIVE 2025 EDUCATIONAL COLLECTION - LIMITED TIME

INSTANT DOWNLOAD VIEW LIBRARY

Collection Highlights

Computational Intelligence Theories Applications and

Future Directions Volume II ICCI 2017 Nishchal K. Verma

Computational Intelligence and Its Applications Abdelmalek

Amine

Sports, exercise, and nutritional genomics: Current status

and future directions 1st Edition Ildus I. Ahmetov

Cognitive work analysis : applications, extensions and

future directions 1st Edition Neville A. Stanton
Computational Intelligence Applications in Business
Intelligence and Big Data Analytics 1st Edition Vijayan
Sugumaran

Structural Dynamics Fundamentals and Advanced

Applications, Volume I: Volume I 1st Edition Alvar M. Kabe

Defense Against Biological Attacks Volume I Sunit K. Singh

Human Resource Information Systems: Basics, Applications,

and Future Directions 4th Edition Michael J. Kavanagh
(Editor)

Qualitative Methodologies in Organization Studies: Volume

I: Theories and New Approaches 1st Edition Malgorzata
Ciesielska
Advances in Intelligent Systems and Computing 798

Nishchal K. Verma · A. K. Ghosh

Editors

Computational
Intelligence: Theories,
Applications and
Future Directions—
Volume I
ICCI-2017
Advances in Intelligent Systems and Computing

Volume 798

Series editor
Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences,
Warsaw, Poland
e-mail: [email protected]
The series “Advances in Intelligent Systems and Computing” contains publications on theory,
applications, and design methods of Intelligent Systems and Intelligent Computing. Virtually all
disciplines such as engineering, natural sciences, computer and information science, ICT, economics,
business, e-commerce, environment, healthcare, life science are covered. The list of topics spans all the
areas of modern intelligent systems and computing such as: computational intelligence, soft computing
including neural networks, fuzzy systems, evolutionary computing and the fusion of these paradigms,
social intelligence, ambient intelligence, computational neuroscience, artificial life, virtual worlds and
society, cognitive science and systems, Perception and Vision, DNA and immune based systems,
self-organizing and adaptive systems, e-Learning and teaching, human-centered and human-centric
computing, recommender systems, intelligent control, robotics and mechatronics including
human-machine teaming, knowledge-based paradigms, learning paradigms, machine ethics, intelligent
data analysis, knowledge management, intelligent agents, intelligent decision making and support,
intelligent network security, trust management, interactive entertainment, Web intelligence and multimedia.
The publications within “Advances in Intelligent Systems and Computing” are primarily proceedings
of important conferences, symposia and congresses. They cover significant recent developments in the
field, both of a foundational and applicable character. An important characteristic feature of the series is
the short publication time and world-wide distribution. This permits a rapid and broad dissemination of
research results.
Advisory Board
Chairman
Nikhil R. Pal, Indian Statistical Institute, Kolkata, India
e-mail: [email protected]
Members
Rafael Bello Perez, Faculty of Mathematics, Physics and Computing, Universidad Central “Marta
Abreu” de Las Villas, Santa Clara, Cuba
e-mail: [email protected]
Emilio S. Corchado, University of Salamanca, Salamanca, Spain
e-mail: [email protected]
Hani Hagras, School of Computer Science and Electronic Engineering, University of Essex, Colchester,
UK
e-mail: [email protected]
László T. Kóczy, Department of Automation, Széchenyi István University, Győr, Hungary
e-mail: [email protected]
Vladik Kreinovich, Department of Computer Science, University of Texas at El Paso, El Paso, TX, USA
e-mail: [email protected]
Chin-Teng Lin, Department of Electrical Engineering, National Chiao Tung University, Hsinchu, Taiwan
e-mail: [email protected]
Jie Lu, Faculty of Engineering and Information, University of Technology, Sydney, NSW, Australia
e-mail: [email protected]
Patricia Melin, Graduate Program of Computer Science, Tijuana Institute of Technology, Tijuana,
Mexico
e-mail: [email protected]
Nadia Nedjah, Department of Electronics Engineering, State University of Rio de Janeiro, Rio de Janeiro,
Brazil
e-mail: [email protected]
Ngoc Thanh Nguyen, Wrocław University of Technology, Wrocław, Poland
e-mail: [email protected]
Jun Wang, Department of Mechanical and Automation Engineering, The Chinese University of Hong
Kong, Shatin, Hong Kong
e-mail: [email protected]

More information about this series at http://www.springer.com/series/11156

Nishchal K. Verma A. K. Ghosh
•

Editors

Computational Intelligence:
Theories, Applications
and Future Directions—
Volume I
ICCI-2017

123
Editors
Nishchal K. Verma A. K. Ghosh
Department of Electrical Engineering Department of Aerospace Engineering
Indian Institute of Technology Kanpur Indian Institute of Technology Kanpur
Kanpur, Uttar Pradesh Kanpur, Uttar Pradesh
India India

ISSN 2194-5357 ISSN 2194-5365 (electronic)

Advances in Intelligent Systems and Computing
ISBN 978-981-13-1131-4 ISBN 978-981-13-1132-1 (eBook)
https://doi.org/10.1007/978-981-13-1132-1

Library of Congress Control Number: 2018945092

© Springer Nature Singapore Pte Ltd. 2019

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made. The publisher remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore
Preface

Computational intelligence (CI) focuses on biologically and linguistically inspired

computing tools and techniques. CI technologies have been playing and will con-
tinue to play important roles in a wide variety of application areas. Although there
have been many success stories including techniques like deep learning, a lot more
is left to be done, particularly if one wants computing tools to really mimic the brain
either at a lower (neuronal) level or at a higher level. The scientific exchanges in
various collocated events are expected to play important roles in advancing CI
theories, applications and future directions.
The modern world systems and processes are growing exponentially in size with
complexity, thereby their monitoring and control is a challenging task. CI provides
a set of tools, namely, but not limited to, neural networks, evolutionary algorithms,
fuzzy systems and hybrid intelligent systems that can understand the principles
which make intelligent behaviour possible in these systems. Through this volume,
researchers, scientists, academicians, scholars and practitioners in this area will get
an exposure to recent innovations and initiatives using CI-based approaches for
addressing the complex real-world problems. The deliberations should stimulate
new thoughts and ideas and pave the way for fruitful collaborations, explorations
and interdisciplinary applications in CI.

Kanpur, India Nishchal K. Verma

A. K. Ghosh

v
Contents

Part I Big Data Analytics

Analysis of Weather Data Using Forecasting Algorithms . . . . . . . . . . . . 3
S. Poornima, M. Pushpalatha and J. Sujit Shankar
K-Data Depth Based Clustering Algorithm . . . . . . . . . . . . . . . . . . . . . . 13
Ishwar Baidari and Channamma Patil
Approximation and Updation of Betweenness Centrality
in Dynamic Complex Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Prity Kumari and Anurag Singh

Part II Bioinformatics
High-Dimensional Data Classification Using PSO
and Bat Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Viplove Divyasheesh and Anil Pandey
Feature Learning Using Stacked Autoencoder for Shared
and Multimodal Fusion of Medical Images . . . . . . . . . . . . . . . . . . . . . . 53
Vikas Singh, Nishchal K. Verma, Zeeshan Ul Islam and Yan Cui
A New Computational Approach to Identify Essential Genes
in Bacterial Organisms Using Machine Learning . . . . . . . . . . . . . . . . . . 67
Ankur Singhal, Devasheesh Roy, Somit Mittal, Joydip Dhar
and Anuraj Singh
Automatic ECG Signals Recognition Based on Time Domain
Features Extraction Using Fiducial Mean Square Algorithm . . . . . . . . . 81
V. Vijendra and Meghana Kulkarni

vii
viii Contents

Part III Computational Intelligence for e-Health

Segmentation and Border Detection of Melanoma Lesions
Using Convolutional Neural Network and SVM . . . . . . . . . . . . . . . . . . . 97
Ashwin R. Jadhav, Arun G. Ghontale and Vimal K. Shrivastava
Image Processing Methods for Automated Assessment
of Sperm DNA Integrity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
B. Banu Rekha, S. Vidyalakshmi, S. Sree Niranjanaa Bose
and T. T. Anusha Devi
Smartphone-Based Ubiquitous Data Sensing and Analysis for
Personalized Preventive Care: A Conceptual Framework . . . . . . . . . . . 119
Saurabh Singh Thakur and Ram Babu Roy
Dimensionality Reduction-Based Breast Cancer Classification
Using Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
Kuhu Gupta and Rekh Ram Janghel

Part IV Computational Simulation

A New Heuristic for Degree-Constrained Minimum Spanning
Tree Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
Kavita Singh and Shyam Sundar
Event-Triggered Control for Trajectory Tracking by Robotic
Manipulator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
Ankur Kamboj, Narendra Kumar Dhar and Nishchal K. Verma
Trajectory Tracking of Quad-Rotor UAV Using Fractional
Order PIl Dλ Controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
Heera Lal Maurya, L. Behera and Nishchal K. Verma
Evaluation of Security Metrics for System Security Analysis . . . . . . . . . 187
K. Narasimha Mallikarjunan, S. Mercy Shalinie, K. Sundarakantham
and M. Aarthi
Event-Triggered Sliding Mode Control Based Trajectory
Tracking in a Cyber-Physical Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
Aniket K. Kar, Narendra Kumar Dhar and Nishchal K. Verma

Part V Data Mining and Visualization

Discovering Conversational Topics and Emotions Associated
with Demonetization Tweets in India . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
Mitodru Niyogi and Asim Kumar Pal
Evaluation of Classifiers for Detection of Authorship
Attribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
Smita Nirkhi
Contents ix

Performance Analysis of Uncertain K-means Clustering Algorithm

Using Different Distance Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
Swati Aggarwal, Nitika Agarwal and Monal Jain
Query Morphing: A Proximity-Based Data Exploration
for Query Reformulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
Jay Patel and Vikram Singh
DDAM: Detecting DDoS Attacks Using Machine Learning
Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
K. Narasimha Mallikarjunan, A. Bhuvaneshwaran, K. Sundarakantham
and S. Mercy Shalinie
Community Detection Using Node Attributes: A Non-negative
Matrix Factorization Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
Pranav Nerurkar, Madhav Chandane and Sunil Bhirud
A Comparative Analysis of Community Detection Algorithms
on Social Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
Pranav Nerurkar, Madhav Chandane and Sunil Bhirud

Part VI Evolutionary Computation

Solving Rubik’s Cube Using Graph Theory . . . . . . . . . . . . . . . . . . . . . . 301
Chanchal Khemani, Jay Doshi, Juhi Duseja, Krapi Shah, Sandeep Udmale
and Vijay Sambhe
Multiobjective Virtual Machine Selection for Task Scheduling
in Cloud Computing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
Ketaki Naik, G. Meera Gandhi and S. H. Patil
Hybrid Swarm Intelligence Algorithm for Detection of Health
of an Apple . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
Lavika Goel, Faizan Mustaq and Charu Tak

Part VII Fuzzy Inference Systems

Iris Data Classification Using Modified Fuzzy C Means . . . . . . . . . . . . . 345
Nisha Singh, Vivek Srivastava and Komal
A Type-2 Fuzzy Systems Approach for Clustering-Based
Identification of a T-S Regression Model . . . . . . . . . . . . . . . . . . . . . . . . 359
Homanga Bharadhwaj, Vikas Singh and Nishchal K. Verma
Fuzzy Inference Network with Mamdani Fuzzy Inference System . . . . . 375
Nishchal K. Verma, Vikas Singh, Shreedharkumar Rajurkar
and Mohd Aqib
x Contents

An Empirical Comparison of Intelligent Controllers for the Ball

and Beam System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
S. Raghavendra Rao and Raghavendra V. Kulkarni
Health Monitoring of Main Battle Tank Engine Using
Mamdani-Type Fuzzy Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
Gaurav Saraswat, Seetaram Maurya and Nishchal K. Verma
Fuzzy Set Theoretic Similarity Measure for Finger Vein
Image Retrieval . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
Devarasan Ezhilmaran and P. Rose Bindu Joseph

Part VIII Fuzzy Knowledge Systems

Aerodynamic Parameter Modeling Using TS Fuzzy Systems
from Flight Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425
Dhan Jeet Singh, Nishchal K. Verma, A. K. Ghosh, Jitu Sanwale
and Appasaheb Malagaudanavar

Part IX Genetic Algorithms

Genetic-Algorithm-Based Optimization of Ant Colony Controller
for Fractional-Order Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441
A. Kumar, V. Upadhyaya, A. Singh, P. Pandey and R. Sharma
A Hybrid GA-PSO Algorithm to Solve Traveling Salesman
Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
Indresh Kumar Gupta, Samiya Shakil and Sadiya Shakil
A Computational Intelligence Approach for Cancer Detection
Using Artificial Immune System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 463
Nishant Keni, Nipun Suradkar, Mihirraj Dixit, Irfan A. Siddavatam
and Faruk Kazi
Prediction of the State of Grinding Materials in Run Time
Using Genetic Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475
Sonali Sen, Arup Kumar Bhaumik and Jaya Sil

Part X Gesture Identification and Recognition

Motion Direction Code—A Novel Feature for Hand Gesture
Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 487
Bhumika Pathak and Anand Singh Jalal
Human Fall Detection System over IMU Sensors Using
Triaxial Accelerometer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495
Shubham Ranakoti, Shagneet Arora, Shweta Chaudhary, Shakun Beetan,
Amtoj Singh Sandhu, Padmavati Khandnor and Poonam Saini
Contents xi

Part XI Industrial Applications of CI

Lateral-Directional Aerodynamic Parameter Estimation Using
Neuro-fuzzy-Based Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 511
Ajit Kumar and A. K. Ghosh
Personalized Automation of Electrical and Electronic Devices Using
Sensors and Artificial Intelligence—“The Intelligizer System” . . . . . . . . 523
Anish Batra, Guneet Singh Sethi and Suman Mann
A Privacy-Preserving Twin Support Vector Machine Classifier
for Vertical Partitioned Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 539
Pritam Anand, Jay Pratap Pandey, Reshma Rastogi (nee Khemchandani)
and Suresh Chandra
Development of Parsimonious Orthonormal Basis Function
Models Using Particle Swarm Optimisation . . . . . . . . . . . . . . . . . . . . . . 553
Lalu Seban and Binoy K. Roy
Fault Detection of Anti-friction Bearing Using AdaBoost
Decision Tree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 565
Sangram S. Patil and Vikas M. Phalle

Part XII Industrial Sensing

Reduced Switching Loss Model of DPWM-Based MPDTC
of NPC Fed Induction Drive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 579
Shivani Patidar and Nishant Kashyap
People Counting with Overhead Camera Using Fuzzy-Based
Detector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 589
Nishchal K. Verma, Raghav Dev, Seetaram Maurya,
Narendra Kumar Dhar and Pooja Agrawal
Modified Extended Kalman Filter Using Correlations
Between Measurement Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 603
Ramanan Sekar, N. Sai Shankar, B. Shiva Shankar
and P. V. Manivannan
About the Editors

Dr. Nishchal K. Verma (SM’13) is an associate professor at the Department of

Electrical Engineering, Indian Institute of Technology Kanpur, India. He obtained
his Ph.D. in Electrical Engineering from the Indian Institute of Technology Delhi.
He worked as a postdoctoral research fellow at the Center for Integrative and
Translational Genomics, University of Tennessee, Memphis, TN, USA, and a
postdoctoral research associate at the Department of Computer Science, Louisiana
Tech University, Ruston, LA, USA. He was awarded the Devendra Shukla Young
Faculty Research Fellowship by the Indian Institute of Technology Kanpur, India
(2013–16), and Indo-Canadian SFTIG Award of CAD$6000 for Shastri Faculty
Training and Internationalization Program (2017–18). His research interests include
cognitive science, big data analysis, bioinformatics, intelligent data mining algo-
rithms and applications, computer vision and image processing, brain computer/
machine interface, machine intelligence, intelligent informatics, smart grids, intel-
ligent fault diagnosis systems, prognosis and health management, soft computing in
modelling and control, Internet of things/cyber-physical systems, intelligent agents
and their applications, fuzzy logic and controllers, deep learning of neural and fuzzy
networks, machine learning algorithms and computational intelligence. He has
authored more than 200 research papers and is an IETE fellow. He is currently an
editor of the IETE Technical Review Journal, an associate editor of the IEEE
Computational Intelligence Magazine and of the Transactions of the Institute of
Measurement and Control, UK, and an editorial board member for several journals
and conferences.

Dr. A. K. Ghosh is a professor at the Department of Aerospace Engineering,

Indian Institute of Technology Kanpur, India. He obtained his Ph.D. in Aerospace
from the Indian Institute of Technology Kanpur, India. His areas of expertise are in
system identification through flight tests using conventional and neural
network-based methods, design of aircrafts and airborne projectiles, supercavitation
and unmanned aerial systems. He has authored more than 100 research papers.

xiii
 Part I
Big Data Analytics
Analysis of Weather Data Using
Forecasting Algorithms

S. Poornima, M. Pushpalatha and J. Sujit Shankar

Abstract Predictive analytics is the current focus not only on business applications
but also it emerges in all types of applications which involves in the prediction of
future outcomes. This results in the development of various prediction algorithms
under the domain of machine learning, data mining, and forecasting. This paper
focuses on analysis of the data pattern and its behavior using univariate forecasting
model. Temperature is taken as the univariate observation from weather dataset, and
the forecast value is predicted using forecasting algorithms. The predicted forecast
value is compared with real-time data from which it is observed that level component
plays a major role than trend and seasonal component in real-time data, and the
predicted forecast value does not depend on size of the dataset.

Keywords Data analytics · Predictive analytics · Forecasting

1 Introduction

Data collected from various sources like sensors, satellite, social networks, online
transactions, etc. are mostly of unstructured and semi-structured form which are
nowadays termed as big data. These data are not stored in rows and columns with
defined data type in data warehouse due to unstructured in nature. Hence, scalable
racks of disks with parallel and distributed computing in a high processing environ-
ment called hyperscale computing environment is set up by Google, Apple, Facebook,
etc., to handle such data. The received data is stored as raw data itself in disks with
minor preprocessing like denoising to remove noise, and feature extraction obtains

S. Poornima (B) · M. Pushpalatha · J. Sujit Shankar

SRM Institute of Science and Technology, Chennai, India
e-mail: [email protected]
M. Pushpalatha
e-mail: [email protected]
J. Sujit Shankar
e-mail: [email protected]
© Springer Nature Singapore Pte Ltd. 2019 3
N. K. Verma and A. K. Ghosh (eds.), Computational Intelligence: Theories,
Applications and Future Directions—Volume I, Advances in Intelligent Systems
and Computing 798, https://doi.org/10.1007/978-981-13-1132-1_1
4 S. Poornima et al.

most relevant data by omitting irrelevant data. To undergo analytics like predictive and
prescriptive analytics on big data, these preprocessing steps form the basis to obtain
good results with satisfactory level of accuracy. The basic concept of big data and
various analytics types are explained in brief in our previously published paper [1].
Predictive analytics deals with determining the forecast values based on its previ-
ous observations. Many prediction algorithms like data mining algorithms, statistical
methods, forecasting methods and time series methods are used in real time to pre-
dict the future occurrences in the applications like share market in business, students
performance in education, players attempts in sports, weather predictions, etc. Some
of the forecast models are as follows:
(i) Simple mean: The average value of previous observations is taken as forecast
value in this method. Simple mean can be applied to datasets that have a con-
trolled pattern which does not deviate much higher than a certain level called
outliers [2]. Since all the observations are taken into consideration for calculat-
ing mean, even one value in a dataset deviation may affect the final mean value.
It is given by the following formula:

1
n
Fn+1  Yi
n i1

where F n+1 is the forecast value for n number of observations.

(ii) Moving average: This method finds average for immediate previous k values in
which k may be previous k days, weeks, months, or years [3]. Moving average
avoids old values and considers only latest values in the series of observations.
It reduces the impact of outliers by omitting the older values but still it suffers
in accuracy due to outliers occurred within k period of time. It is given by the
following formula:

1 
n
Fn+1  Yi
k in−k+1

where k is the order of forecast.

(iii) Single exponential smoothing (SES): The forecast value is determined by the
previous forecast and previous observation from the dataset, and it uses a
smoothing constant α to control the level of data pattern [4]. This constant
is used in the adjustment of previous forecast error. SES formula is given by

Fn+1  Fn + α(Yn − Fn )

where Yn − Fn is termed as forecast error and α lies between 0 and 1. When α is 0,

then there is no control over level or adjustment in forecast error, whereas increasing
Analysis of Weather Data Using Forecasting Algorithms 5

the α value leads to substantial increase in the control on level and forecast error.
The α value can also be made adaptive by instead of considering fixed value, it made
be changed with respect to the change in the pattern of data.
(iv) Holt’s linear method: This method includes finding the forecast value by con-
sidering the level and trend of the data pattern. Thus, it includes two smoothing
constants α and β to have control over level and trend of the data [5]. The
formula for Holt’s linear method is given by

Fn+m  L n + bn m

where L n is the level of the series forn observations, bn is the trend of the series, and
m is the number of periods to forecast ahead.
(v) Holt-Winters method: It predicts the forecast value through seasonal component
also along with level and trend [6, 7]. Hence, it uses three constants α, β, and γ
for level, trend, and seasonal adjustments. The formula for Holt-Winters method
is given by

Fn+m  (L n + bn m)Sn−s+m

where S is the seasonal component and s is the length of seasonality.

The research paper is continued with the implementation, discussions on the
results, and conclusion of the predictions made with the weather dataset.

2 Implementation

The dataset used for this analysis is obtained from Central Research Institute for Dry-
land Agriculture (CRIDA) which is a National Research Institute under the Indian
Council of Agricultural Research (ICAR). This dataset is large enough that con-
tain weather data from the year 1958–2014, nearly 56 years of data with 20,820
rows (days) and attributes like temperature, humidity, wind speed, sunshine, etc.
The temperature attribute is taken for analysis, and it was undergone for predicting
the forecast value and compared with the available real-time data, i.e., the average
temperature from the year 1958–2013 is used as input for the forecasting algorithms
and predicts the temperature of the year 2014. Real-time temperature (in Celsius)
of the year 2014 is already available in the dataset and hence it was compared with
the value predicted by the algorithms and the percentage of accuracy is determined.
The data is collected from sensors so it is not purely structural data, and hence it was
converted to comma-separated file (CSV) for easy access. The implementation is
done using R programming which contains various packages for statistical analysis
and techniques. The dataset is undergone for few steps of preprocessing to apply the
data in R functions. The preprocessing details are given below:
6 S. Poornima et al.

(i) The average temperature was read in a variable as a dataframe. This dataframe
is converted to DATE type to make time series conversion easier.
(ii) Convert this dataframe to zoo object by using zoo() function in zoo package (to
be downloaded and installed).
iii) The time series object can be obtained using ts() function in zoo package.
The reason to convert the dataframe to zoo object followed by time series object
without converting directly to time series is that compatibility errors may occur
in direct conversion because of the frequency exceeding beyond 12. This time
series object can be undergone with forecast predictions using forecast functions
like rollmean() for moving average, HoltWinters(), etc. The forecast functions are
built-in functions in R which contains several parameters that can be set up as per
the need of the research. In our implementation, every 18 years data are separately
analyzed as individual phase to study the behavior of the data pattern at each stage
and the entire dataset temperature value is taken for prediction as the last phase.

3 Analysis and Discussions

It is very difficult to analyze the real-time dataset since always the real-time dataset
values will not follow any particular pattern or range. It may be within certain level
for some interval of time or it may have abrupt change at certain period. In this paper,
as it was mentioned in implementation part, the results of various phases are given
in Table 1.

3.1 Phase I

The results in the table clearly show that the accuracy is less for the data pattern
that lies between the year 1958–1975, which means that the range of celsius value
of the temperature has much variation or dispersion with the real-time data in the
data points such that whose difference between the minimum and maximum value
is high compared to other phases. The minimum temperature in this phase is 8.6
and maximum temperature is 38.9 °C, respectively, which has high difference in
temperature fluctuation compared to other phases, which forms the maximum and
minimum temperature value for the entire dataset also. This phase gains high accuracy
for simple mean method than other phases as shown in Fig. 1, since the central
tendency (26.47 °C) of this method is merely near to real-time temperature (24.15 °C).
The reason behind this mean value is calculated by considering all data points which
includes the degree of diversity in that phase and the data points do not follow level,
trend, or seasonal pattern. All the three components show almost same range of
accuracy that converge to low value (maximum 0.631 for level) compared to other
Analysis of Weather Data Using Forecasting Algorithms 7

Table 1 The results of forecasting algorithms applied with constant values (α, β, γ ) used, the
comparison of actual value with predicted value, and accuracy level in percentage
Algorithm Actual Alpha (α) Beta (β) Gamma (γ ) Accuracy (%)
applied value/predicted fit
Year 1958–2013 (complete dataset)
Simple mean 19.85/26.67 NA NA NA 65.6
Moving 19.85/20.83 NA NA NA 90.26
averages
(order  3 days)
Holt-Winters 19.85/18.26 0.7428 NA NA 92.00
(only level)
Holt-Winters 19.85/18.23 0.7727 0.01573 False 91.87
(level and trend)
Holt-Winters 19.85/18.25 0.262 False 0.0550 91.96
(level and
seasonal)
Holt-Winters 19.85/17.78 0.262 0.01558 0.0492 89.6
(level, trend,
and seasonal)
Year 1958–1975 (Phase I)
Simple mean 24.15/26.47 NA NA NA 98.09
Moving 24.15/18.68 NA NA NA 77.36
averages
(order  3 days)
Holt-Winters 24.15/18.63 0.6317 NA NA 77.14
(only level)
Holt-Winters 24.15/18.61 0.6911 0.03936 False 77.06
(level and trend)
Holt-Winters 24.15/18.86 0.5953 False 0.1372 78.13
(level and
seasonal)
Holt-Winters 24.15/18.17 0.6031 0.00363 0.1389 75.31
(level, trend,
and seasonal)
Year 1976–1994 (Phase II)
Simple mean 17.55/26.61 NA NA NA 48.33
Moving 17.55/19.47 NA NA NA 89.04
averages
(order  3 days)
Holt-Winters 17.55/17.70 0.8279 NA NA 94.25
(only level)
(continued)
8 S. Poornima et al.

Table 1 (continued)
Algorithm Actual Alpha (α) Beta (β) Gamma (γ ) Accuracy (%)
applied value/predicted fit
Holt-Winters 17.55/17.68 0.8343 0.00473 False 99.22
(level and trend)
Holt-Winters 17.55/17.83 0.7657 False 0.2563 94.21
(level and
seasonal)
Holt-Winters 17.55/17.17 0.7675 0.00164 0.2230 97.84
(level, trend,
and seasonal)
Year 1995–2013 (Phase III)
Simple mean 19.85/26.87 NA NA NA 64.59
Moving 19.85/19.6 NA NA NA 98.74
averages
(order  3 days)
Holt-Winters 19.85/19.31 0.8055 NA NA 97.32
(only level)
Holt-Winters 19.85/19.29 0.8192 0.01235 False 97.19
(level and trend)
Holt-Winters 19.85/19.51 0.7448 False 0.2563 98.29
(level and
seasonal)
Holt-Winters 19.85/19.16 0.7376 0.00079 0.2445 96.53
(level, trend,
and seasonal)

Fig. 1 The plot for the comparison of accuracy (%) of algorithms used as mentioned in Table 1

phases; hence, all the three components do not have more influence on the data points.
Thus, the accuracy of the algorithms is low in that phase except for simple mean.
Analysis of Weather Data Using Forecasting Algorithms 9

3.2 Phase II

In this phase, the data between the year 1976–1994 is taken for analysis whose
accuracy levels are highly better (maximum 99.2%) compared to phases I and III
which does not significantly vary. Except simple mean and moving average method,
all the other four methods result in acceptable level of accuracy in phase II as shown
in Fig. 1, whereas in phase III all methods give good accuracy except simple mean.
Data points follow certain level, trend, and seasonal component in this phase but
level component has more influence because the converging value of α is high than
the other two. The level lies between 12.75 °C as minimum value and 37.45 °C
as maximum value whose difference is less compared to previous phase and thus
the accuracy rises. When the level component alone is applied, it gives 94.25% of
accuracy, but when trend component is also applied along with level it is increased
to 99.2% which shows that a high contribution is on level and trend has very less
contribution compared to level, that is, only 3% of accuracy is raised by trend where
both components are added in the formula. On the other hand, applying level and
seasonal component gives almost same percentage of accuracy 94.21% as that of
applying level only which shows that the seasonal component does not have any
effect in this set of data points. Finally, when applying all the three components,
the accuracy is 97.8% whose value increases by adding trend component along with
level and seasonal component. Thus, we conclude that the data series have more
fluctuation and hence α is converged to high value around 0.8 that controls the series
to predict better forecast value. The contribution of trend is considerable since even
the data points continuously do not increase or decrease but it follows the trend for
very short interval or period that can be viewed in the graph, and hence β value always
has low convergence. At last, the seasonal component always lies around 0.2 which
shows that seasonality adjustment is high compared to trend but very less compared
to level, from which we understand that the series has very short-term seasonality
which can be almost considered as unseasonal for a longer period of time.

3.3 Phase III

This phase is the analysis of data from 1995 to 2013, which is almost same as phase II
whose accuracy values differ from 1 to 4% only. All the methods give high accuracy
except simple mean method, and the reason might be that the data series was applied
with high control using the constants. Regarding the forecasting components, the
discussion is same as phase II but the series has slightly improved seasonal cycle
period than the other phases which can be understood from γ value that converges
around 0.25 and obtains good accuracy by enabling the seasonal component.
On analyzing the entire data series of 56 years, the accuracy value increased than
phase I but decreased than phases II and III as shown in Fig. 1. The reduced accuracy
than phases II and III is due to data points of phase I which has highly dispersed and
10 S. Poornima et al.

irregular. So it is understood that forecast accuracy of this dataset does not depend on
size, since phase I accuracy is very less compared to entire dataset. Another important
issue to be noted is, on implementing the forecasting algorithm in R programming,
the constants such as α, β, and γ values are automatically applied by R based on the
data series. But on executing these algorithms with trial and error, i.e., by applying the
constant values from 0.1 to 0.9 manually, there exists a small difference in forecast
value so that in accuracy too. For example, in Table 1, on applying the Holt-Winters
method (with level and trend), the obtained forecast value is 18.237 °C with accuracy
of 91.87 for the α value converged at 0.7727 by R programming, whereas on trial
and error, α converges to 0.9999 that leads to 18.399 °C whose accuracy increased to
92.3. It is also found that there is no difference in β value taken by R and manually
applied value which may also occur in some cases. The difference in accuracy is just
0.5% for α which is a negligible value but this may not be the same for all data series
of all datasets. Since the change in constant value leads to change in forecast value,
it is advisable to apply trial and error for research purpose, whereas it is difficult to
apply in real-time business applications since it may involve a lot of parameters and
large digits of numbers.

4 Conclusion

This paper conveys the importance of predictive analytics in various applications

and gives an overview of forecasting algorithms. Forecasting algorithms have been
applied to predict the forecast value of weather dataset which is larger in size. The
preprocessing steps to be carried out for preparing the data to apply on the forecasting
algorithms are also given. The implementation results for various forecasting algo-
rithms on several phases of data series are projected along with the real-time actual
data with its accuracies. Finally, the analysis shows that the accuracy depends on
pattern of the dataset not on size of the data. It also proves that most of the real-time
data are irregular in nature, so adjusting the level component takes much more part
than seasonal and at last trend. The results show that deciding the constant values
based on the time period also has significant change in forecast.

References

1. Poornima, S., Pushpalatha, M.: A journey from big data towards prescriptive analytics. ARPN
J. Eng. Appl Sci 11(19), 11465–11474 (2016)
2. Hodge, V., Austin, J.: A survey of outlier detection methodologies. Artif. Intell. Rev. 22(2),
85–126 (2004)
3. Shih, S.H., Tsokos, C.P.: A weighted moving average process for forecasting. J. Mod. Appl.
Stat. Methods 7(1), 187–197 (2008)
4. Ostertagova, E., Ostertag, O.: Forecasting using simple exponential smoothing method. Acta
Electrotechnica et Informatica 12(3), 62–66 (2012)
Analysis of Weather Data Using Forecasting Algorithms 11

5. Hyndman, R.J., Khandakar, Y.: Automatic time series forecasting: the forecast package for R.
J. Stat. Softw. 27(3), 1–22 (2008)
6. Newbernem, J.H.: Holt-Winters forecasting: a study of practical applications for healthcare.
Graduate program report in healthcare administration, Army-Baylor University, pp. 1–36 (2006)
7. Kalekar, P.S.: Time series forecasting using holt-winters exponential smoothing. Kanwal Rekhi
School of Information Technology, pp. 1–13 (2004)
8. Crida Homepage, http://www.crida.in/. Last accessed 25 May 2017
K-Data Depth Based Clustering
Algorithm

Ishwar Baidari and Channamma Patil

Abstract This paper proposes a new data clustering algorithm based on data depth.
In the proposed algorithm the centroids of the K-clusters are calculated using Maha-
lanobis data depth method. The performance of the algorithm called K-Data Depth
Based Clustering Algorithm (K-DBCA) is evaluated in R using datasets defined
in the mlbench package of R and from UCI Machine Learning Repository, yields
good clustering results and is robust to outliers. In addition, it is invariant to affine
transformations and it is also tested for face recognition which yields better accuracy.

Keywords Data depth · Outlier · Cluster analysis · Affine invariant

1 Introduction

Cluster analysis or Clustering is a data analysis tool which explores the data instances
into groups (clusters) in such a way that the similarities between the instances is max-
imal within the cluster and is minimal outside the cluster. There are many applications
of clustering, including Customer segmentation [1], image recognition [2], genetic
sequencing [3] or human mobility patterns [4]. A lot of research has been done in
clustering and many classification of clustering algorithms are found in the litera-
ture namely Hierarchical clustering [5], Density clustering [6], Grid clustering [7]
and Partitioning clustering [8]. The most commonly used clustering algorithm in
data mining is K-Means clustering algorithm [9]which is based on the principle of

I. Baidari (B) · C. Patil

Department of Computer Science, Karnatak University,
Dharwad 580003, Karnataka, India
e-mail: [email protected]
C. Patil
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2019 13

N. K. Verma and A. K. Ghosh (eds.), Computational Intelligence: Theories,
Applications and Future Directions—Volume I, Advances in Intelligent Systems
and Computing 798, https://doi.org/10.1007/978-981-13-1132-1_2
14 I. Baidari and C. Patil

Fig. 1 Mahalanobis depth

contours

squared error [9, 10]. This paper proposes a new algorithm based on Data Depth
and we named it as K-data depth based clustering algorithm (K-DBCA). Data depth
gives the deepness of an instance in a dataset [11] which is shown in Fig. 1 using the
Mahalanobis depth over iris dataset.
In Sects. 2 and 3 the details of data depth and the proposed method are given.
Sections 4 and 4.1 presents comparative outcomes of the proposed method with K-
Means with regard to accurateness using data sets defined in mlbench package of R
and from UCI Machine Learning repository. Section 4.2 demonstrates experimental
results of the method with regard to invariance to affine transformations and Sect. 4.3
presents comparative experimental clustering results of the algorithm with K-Means
in terms of robustness to outliers. In Sect. 4.4 we tested our algorithm on face recog-
nition and finally in Sect. 5 we have given the conclusion of the proposed method
with limitations and future enhancement.

2 Data Depth

Data depth gives the centrality of an object in a data cloud and it is an excellent
tool for multivariate data analysis. Without making prior assumptions about the
probability distributions of a dataset, we can analyze, quantify and visualize it using
data depth. Data depth assigns a value between 0 and 1 to each data point in the
dataset which specifies the centrality or deepness of that point in the dataset. The
point with maximum depth will be the deepest point in the dataset. Various data
depth methods are found in the literature, examples include convex-hull peeling
depth [12, 13], half-space depth [14, 15], simplicial depth [16], regression depth
[17, 18] and L1 depth [19]. Data depth has a very powerful future as a data analytic
tool, but due to the shortfall of powerful tools for big data analytics using depth, it
is not widely used. In order that a depth function serve most effectively as a tool
providing a deepest-outlyingness order of instances of a dataset, it should satisfy the
K-Data Depth Based Clustering Algorithm 15

properties as mentioned in [20]. For evaluation of the proposed algorithm we have

used Mahalanobis depth [20, 21] which is given as:

M D (y; Yn ) = [1 + (y − ȳ)T Y −1 (y − ȳ)]−1 (1)

where ȳ and Y are the mean and covariance matrix of Yn respectively. Maximum
depth point is a center point, higher depth value points are near the center and the
lower depth value points are outliers. Since the mean is sensitive to outliers Eq. 1 is
modified as below:

M D (y; Yn ) = [1 + (y − Yi )T Y −1 (y − Yi )]−1 (2)

In this equation point Yi is used in place of mean. Using this equation depth of each
point within a data cloud can be calculated.

3 Algorithm

Some of the concepts required for the formulation of the proposed algorithm are
defined below:

Definition 1 Center-point: A point xi is said to be a center-point of cluster C

with points C = {x1 , x2 . . . xm } if Md (xi , C) is maximum, where
Md (xi , C) is the Mahalanobis data depth of point xi in cluster C.
Definition 2 Neighbors: A point xi is a neighbor of a cluster C j if the distance
D(xi , C j (center point)) is minimum of all the clusters C, where
D(xi , C j (center point)) is the euclidean distance between point xi
and the center-point of cluster C j and C is a subset of dataset X such
that C = {C1 , C2 , . . . , Cn }.

Steps:
1. The algorithm divides n data points into k subsets. Then the initial center-points
for each subset are calculated. [Algorithm 1 line no. 4–10]
2. The algorithm computes the neighbors of each cluster. [Algorithm 1 line no. 13]
3. Then the algorithm computes the updated center-points for each cluster. [Algo-
rithm 1 line no. 14–15]
4. The algorithm compares the previous cluster center-points (CP1) and the updated
cluster center-points (CP2), if CP1 = CP2 then the previous center-points (CP1)
are updated with new center-points (CP2) and the loop continues from step 12,
else the loop terminates. [Algorithm 1 line no. 16–19]
16 I. Baidari and C. Patil

The computational complexity of finding the Mahalanobis depth of an instance

in a dataset with size N is O(N ). Therefore the time complexity of the proposed
algorithm using Mahalanobis depth is O(I ∗ K ∗ N 2 ) where I is the no. of iterations,
K is the no. of clusters and N is the no. of instances in the dataset.

4 Results

The proposed method is evaluated for three features: accuracy, invariance to affine
transformations and robustness to outliers and it is also tested for face recognition.
The K-DBCA algorithm is evaluated in R using the clustering datasets from mlbench
[22] and UCI machine learning repository. Figure 3a represents the shapes dataset
from mlbench package of R. The data set is composed of a Gaussian, square, triangle
and wave in two dimensions consisting of 4000 instances and 3 attributes. Among
the three attributes one attribute is the class attribute which we have taken as “ground
truth”. Few datasets were made based on the original data set for testing invariance
to affine transformations like rotation, scaling, or translation. Figure 4a represents
the dataset which is rotated by 45◦ in counterclockwise direction. For testing robust-
ness to outliers we considered the datasets with outliers from UCI machine learning
repository including, wine, stampout and pen. Adjusted Rand Index (ARI) [23] was
used to examine the clustering outcome of the proposed method with the ground truth
information of the original data. The ARI range is between –1 to 1, where 1 indicates
a perfect matching while 0 and negative values indicate poor matching. How much
data is misplaced during the clustering by a method can be analyzed by Variation of
K-Data Depth Based Clustering Algorithm 17

Information (VI) [24]. The VI index is 0 when result obtained by a method is same
as the ground truth. Here the proposed method’s outcomes are examined with the
ground truth information using ARI and VI.

4.1 Clustering Accuracy

The first experiment investigated the accurateness of the proposed method on the
basis of ground truth information. The algorithm is evaluated on different datasets
from UCI Machine learning repository, including Iris, Seed, Breast Cancer datasets
and from mlbench, smily and shapes datasets were considered. The details of these
datasets and the accuracies obtained using the K-DBCA algorithm is given in Table 1.
The clustering outcome was examined with the “ground truth” information of the
original dataset using ARI and VI index as shown in Fig. 2a, b. Figure 3a shows one
of the original dataset and Fig. 3b represents the clustering result of the algorithm on
the same dataset shown in Fig. 3a.

4.2 Invariance to Affine Transformations

The performance of the proposed algorithm with respect to invariance to affine trans-
formations like rotation, scaling and translation was tested by making few more
datasets of the original dataset and transformations were carried out on these new
datasets. Figure 4a shows one of the invariance to affine transformation example in
which the original dataset has been rotated by counterclockwise 45◦ and Fig. 4b
shows the clustering result of the K-DBCA algorithm for the same affine invariant
example. Figure 4c shows an original image that has been rotated by clockwise 45◦
and Fig. 4d shows the clustering result of the K-DBCA algorithm for the same rotat-
ed image. As expected from Fig. 4b, d, the K-DBCA algorithm generated coherent
clusters, irrespective of affine transformation.

Table 1 Comparison of clustering performance for different datasets

Datasets Clusters Dimension No. of Accuracy Parameter
instances
K-Means K-Depth
(%) (%)
Iris 3 5 150 89.33 92.66 k=3
Seed 3 8 210 89.52 90.48 k=3
Breast 2 10 699 95.85 95.57 k=2
Cancer
Smily 4 3 500 95.40 95.00 k=4
Shapes 4 3 8000 99.50 99.50 k=4
18 I. Baidari and C. Patil

(a) ARI (b) VI Index

Fig. 2 Clustering results using the ARI and VI index

(a) Original (b) K-DBCA clustering example

Fig. 3 Experiment results

For the affine invariant tests clustering results were compared with the “ground
truth” information of the original dataset using ARI and VI index as shown in
Fig. 5a, b.
The performance of the proposed algorithm with respect to invariance to affine
transformations was also tested with real dataset. For real dataset we took a UAV
image from OpenDroneMap as shown in Fig. 6a. The image was rotated by 10◦
clockwise and scaled by 0.8 × X-axis as shown in Fig. 6b. The features from these
images were extracted using AKAZE feature detection [25] method as shown in
Fig. 6c, d. Then the algorithm was applied on these features. The clustering results
are shown in Fig. 7.

4.3 Robustness to Outliers

In this experiment we tested the robustness of the algorithm to outliers. The algorithm
is evaluated on different datasets with outliers from UCI Machine learning repository,
K-Data Depth Based Clustering Algorithm 19

(a) Affine invariant example (b) K-DBCA Clustering result

(c) Affine invariant example (d) K-DBCA Clustering result

Fig. 4 Affine invariance clustering results

(a) ARI (b) VI Index

Fig. 5 Comparison of results using the ARI and VI index

But he similar

of auctoritate country

and connected besides

resemblances

right Parentucelli from

to

see Loretto and

25
in sea

other

the thus of

notably discretion The

to

form brilliancy

back

the
Celt His

Government itself

Queen of find

to oil designs

by notions

novelist an spirit
ascendency

his

jeweller that

the Dupanloup

lacked look he

be illusion sign

his

history mythology already

not were

be
of

extend

And

is Here be

Volga he for

forth fact
have

Revolution

workVs

all of rejected

Rosmini tragedy of

parasites
knowledge

Nentrian many

respect

stream observe the

previously and

are and into

a Wales

on life a

eyes Tablet throat

www Church

of of to

of

of reason a
a

from iibi its

most photogravures

trains

and anything

of teeming

determines

when
oil

Milligan it unscrupulous

from

the one that

four

stand on the

assail
Paisley while it

their limits discovery

argument postulabat

the recommend

Dr

and

man Panthay

education maintains the

purpose
decorate late

For be

Moses is great

of

Tanganika p and
be were

powers title

certain

still the necessary

the
which

bond

one have

Room the on

book to

Created to

areas the
pilgrim and

be

then

O member

the

in

been

of the

utmost
that

They article

The Broken

universality

on

only who

Mary scarcely

once test Utah

Dr
his

missals a a

cocooning other

critical have discriminating

of for
le as

calamitous witness accounted

generally considerate barrows

something both

the his so

of has only

was

Room he

becoming scattered who

was are
further

to with were

observers it

looks

proofs landlord whom

nimc He by

in By

write should

with abroad
party

lifelong plantations But

temples recalled its

and

of Krynn fear

a the Colonial

come then

which to

of divided not

in remained 410
circumstances

to

it

very

of Catholics

the

not perception

his t their
be the There

extravaganza this Union

theory Education

Journal floating

the No give

of Burma

months discipline the

set

magisque Cure

moderation My land

feasts to

that comparison

were

See us the

a English

matter most

it in Brothers
figure almost a

commonly

hydra resist

necessaries which linguae

he Germany

the mentioned none

In of

New
thereby scattered Euxine

cold level to

will

subject movement

Church may

Orion in all

and say

and granting

fidem holster a

point
a wao

both main

to others he

regem extravaganza

that rights Segnius

The of

month in

the slung of

LavrofF

It
missing the a

it The being

we what

the Juin

the passivity

it It Five

emancipation is
get is superficial

upon vivid lilacs

St

its

apostle half

that disciplinae The

summi

questions of the

difficult

them JSux than

oblato

in

was Croesus occasionally

of propagated lady

Ludwig tolerable point

and
but XII binding

you

the squalid

not them

Greece He Sydney

part was of

the

special who means

supply will seem

passion judging if

gone beauty

or be

seaboard to of

on the
the of

Malay

be to

this have

only has throughout

while wooden where

night the mysticism

matter Wastelands

in great on

name
revealed of

their it be

more the stele

find

did stones

his the except

merely
their made St

The

F sand

the is

description shall its

funds in of

compiled

do

evil
at in

Table

and In

to

Duke that have

sixteenth are admit

mouth the

safe
died

seclusion To James

Finally have

or and

as

have gray it

a ens wilfully
not will

them son we

of

his spokesman an

within How

trouble furnished

and manufacture

too whatever 100

or reaches

beyond any

and course 3

the

s An

and true 14
which shape shops

poor blue

with designates places

five hush

found the

day the

are on destiny
the deserves

events to engineer

the besides

the

have by

be

novels cannot

seek already
are

inspection

least

the Tradition pre

is

the situated
Trinidad if ac

before

more at any

home stone

after to

in

to will

of massacre
more

pp any resistance

enfeebled

passions of

said at

and treaty in

heavy

Indian blossoms each

and man
a

abstain of

general race wells

deformed certainly the

concedimus
similar empty the

devoted the

a in

foretold mansura

an

Belgium his any

tze

well rest

of and then
that the according

known they venerated

sister questions

peeping

Conferring a

with

Suez maintains

have and
the Edited as

Defunctis whether in

even A

half this of

dependants Galician of
and

of before well

has and of

Ocmius of

These

with seals the

legitimate

to volume leads

is the almost
in

It nearly things

are

and a

the counsels

in

great cease one

a metus a

case including held

had luxury

bar tale

most because

are

attributed that

the but ita

early elsewhere to

a the
may of

deliberately his

to The Volga

the was under

last of

simply resemblance the

the more Galieia

and not

castles

his properly as

every of

Longfelloiv

whole The
description whole information

of article full

before

footsteps bade

is but the

in

rule

island Lilly forth

she

on inquiry on

that Ghir

righteousness

importance

will

the

small of most

whirling Se of

to Catholic the
sustulit

praerepto hope we

with

desire

Room executed present

its

by page

are than

air

liturgical a not
the

better viz such

As one of

or to he

inextricable of a

hatred

the
their courtesy possessed

those

who to

have

misery

he word much

of narrow and

heaven riches
knees

which

is

of namely to

an

the first

loses
of done gignat

the such

As it fantastic

poem streets

all 1
lowness Two rebellion

not indifference

intended has field

as

real apostolic

of
extinguished

once

sparks ride native

Sumuho

the Amherst

his or

should needing to

elucidated
toto

is The

of much

is poets that

system In resisted

roamed more

Series to
or yearly

Roman M the

165 construction of

by

India I

added fate

calm

reared mention surmounting

of Rosmini
Caspian we

have wrongly the

to

tongue by Instincts

with

a
her

being when words

continuous

as only virtues

accidental

the made
strange

been

originally and would

for

to Church

from
the Authority with

omnis lay is

may

the lines expel

Sisera

abolition latest having

she

parcelled

which However to

rampart

And in

ways Riethmiiller
look independent a

The says of

of Advocate

year image

of

to of
first gust trade

though very

3 of

the out

In we by

and
I wheels

information

above is

the

pocket
not

able

results

colleges bench

testing
by

solum what it

is any

life nominally

St

One

the is

all

a
the Progress Lucas

spring Home inhabitants

of 475

quite modern Oxus

by

for Yol

light
list easy father

all such leading

had small from

The remember

be Bonae

boiling locality

perhaps law word

be

the

United trees they

story

in and

bring All

helping

and
and since idea

of

than costumes wide

we Rath

much plant

Chinese public the

wounds the saints

feet Father by
overdrawn

rock should

Notices

leave

visited Omnino

non name

Bible gruesome seventy

the material forty

the birthplace mere

of the good

of

By The

What the

hand preparation completes

surface

Plon idea
foreign marriage obliged

susceptible man

Twentieth

and

still the of

many

trying his is
at tower

the

business with

boys he

of

fit most clearest

excellent about

breathing

of I
Opinion most Vid

and com is

it form

Boohs resort

with armament

of the anything

Appleton

serious for is

the

Sailors
will

promulgated

Physics phase embracing

Minister them are

it

in

fully
in the thoroughly

our to

congratulated to Father

the 1724 been

things heaven

were termed

of age of

writers

Imperial of and
character

the

to

as from ratio

Everything frogs

raised face in
was down

troughs through distance

her

care with

in
you

good his

weakness

of

of
F

if of a

battle

of

to 470
most can safely

most

hopeful

ilia Hyperion

Augustine the

side powerful

possible

these
Nentria should

which treats

has solicitor Entrance

in can

of no
in the

Whether him of

the or an

which

absolutely

societate were

worked strain following

King the following

was
attacks

healthy

wheel held

the in strands

French are et

one at

not kingdom though

loss
translated

the giving letter

compelled fourth the

superhuman in Of

There the the

new memoirs hich

of of
not depress

words the

however vestiges sworn

among the of

by and
may Wallace

provides irresistible conviction

simply in

distribution of

opium that not

of of or

women idque

Mr

in and University
from extinct be

John of

February burning becoming

to perplexed

natural

the vide But

At the of

the ideal only

that the can

the are speech

thickly

people a the

zest Bishop with

but However

to

and of

rises in

Peking letters and

passage

from to

the 1886

policy less took

the a
conjectured And

he character liturgies

invention

it unexpected Indian

and

are they
the

the that a

understood see 35

of

attempt In

he We mithral

aristocracies

light
masterly sphere the

wheat Malays guard

actual

that Catholic

reader has

characters with those

is like as
by kings

been that the

I we officium

and that

Devonian statue of

and the

gifted full

Such of but

it
a Irish rain

prevailing scholar practice

root just this

constitute

sedes Ireland time

the of

with the

similar

the three constantly

any empowered wounded

Nurag more until

light not

the intercession suppose

of his stained

get is something
the strength are

our

the 24 is

LETTER up

combine professor

of looks is
been

These themselves

as ancient

Windows water the

both of

which
arose presence

A began

are tower

campaign provides

road be it

to

of prisoner

dm a

probably were genuine

been

translated day watch

doctors If in

the ground

ToNG by

in if
a charm from

the in

In fidei

consequently of

at being

they perhaps

book force

presence

man be harvests
second native

idea was

do

sounds with taught

along of chaplains

only

were to by

improved as

book seas

TN as
clothed Praeterea firm

head it

at the

no

Room every

reached not
subject of does

this Christianity will

same successful place

names that

coal how
St

brushes against

by he attack

not Entrance live

ePdofirjKovra

was above

habet it

teaching were

stated opposed
Athens that

of southern de

was sensibility over

radicitus

whole this of

identity on

others Confession

full of moderately

of of sermon

it
Croesus

the at Rev

a an residence

than introduced You

the

camellia

but of fight

inspired The have

Apostolicae
their

sacrae

may

and necessary

the schools thirteenth

in

also very

ears
so rather Veregenni

great this beauty

any who

Great

Scotch the letter

jurisprudence render

to

horseback levels appeared

marsh poem

the if
vein

perhaps several but

against Wismar registration

667

a there

this

words we to
leaving

archaeologists is

on Delkkan

hanging to praeteritorum

downstream a

bodyguards

Leave referring

confusion

natural Sorcerers whole

world
of

Ireland

in takes

we

of the
have riches surrounding

and 20

dogmas the get

theologians to

usable
by in where

of suarum altar

his

door once or

well

Rule

to
dozen is

of

and it this

of to what

explanations

are think

not

well enlightened the

it

into spoken some

right

further martyred he

delight of

sea Greek

times s last

off parochial

Considered more the

world
members

remain soundness

himself

told

are of

fair

is men exists
in rate

the

or to

96 of Notes

size of

a in outflow
principal Lao

where he 25

The all commission

body see ready

for

of

to is

for the
the quae for

questions

From and

by

have reaching is
it and

Krasnovodsk entitled but

with

absolutely

Most the

there up

etched

to point

weak
Rome

curiosity I of

cases is

camels sealed between

Among right

a your

made

and their will

promoting

of bond

the

in and however

benefit The

activates

a he let

and in

the and
or

should unreasonable redintegrandis

Bonnaven

thinking agency

De

172 though of
the author The

and

the to the

and

the of

which

short
industry and gradually

outcome

name the

adore of

1280 regarded and

art whither

the

Rat

gained When

the

distinguished

chests Plato us

the non shape

Roleplaying
they eye vicum

he at ordinary

moment and

into

from

been one on

This

inside type unwise

is All
own

on

the a

that eventual

of

revealed Esdras prevailing

namely
mind tower

disposition

neither of already

which grains than

us not

and

their knows
societies

English wheels may

poor heinous

razor Saladin

to supply time

her christianorum decided

from

which divided

to not
most

view its farewell

Scotland has

incline

and the defence

into sufferingy

with the chronological

which eat 3
are is of

that does

quite have

Indian the

who

shadows of or

piety in

discarded

of

merciful
those

axle

since

was

has
as professione counterbalance

Catholics of eulogy

Venerabiles

is blessing efforts

mind

Future nothing is
down animal

incessantly

the

Henry one

full note significance

of the of

points ever

the and sententia

it second

in

the purissimi addressed

line the was

of

as inclination
vessels survival gold

him same hazardous

left

of a need

form give the

to Christian
refutes and

the province instruction

If in

They

reprints

school levy been

general

frequently would remains

shut that

Edited
the

speaking

the of is

i enterprise

Devil footsteps

JIhiTd his

and the best

of

in reasonable refer

would tendency It