UNIT-IV

Object Segmentation: Regression Vs Segmentation – Supervised and Unsupervised

Learning, Tree Building – Regression, Classification, Overfitting, Pruning and
Complexity, Multiple Decision Trees etc. Time Series Methods: Arima, Measures of
Forecast Accuracy, STL approach, Extract features from generated model as Height,
Average Energy etc and Analyze for prediction

Regression Vs Segmentation:
Regression:
Regression analysis is a form of predictive modelling technique which investigates
the relationship between a dependent (target) and independent variable (s) (predictor). This
technique is used for forecasting, time series modelling and finding the causal effect
relationship between the variables. For example, relationship between rash driving
and number of road accidents by a driver is best studied through regression.
Regression refers to average relationship between two or more variables. One of these
variables is called the dependent or the explained variable and the other variable is
independent or the explaining variable. Regression is one of the statistical method that is used
to estimate the unknown of one variable from the known value of the related variable.
For Example regression helps finance and investment professionals and other business
professionals. It helps in predicting company sales depending upon weather, previous sales,
GDP growth etc., Regression considers a set of random variables to find the mathematical
relationship between them. The relationship is mostly in the form of straight line which
approximates the separate data points. There are two types of regression namely simple linear
regression and multiple linear regression.
Simple linear regression represents the relationship between two variables where one of them
is independent variable X and other variable is dependent variable Y.
Multiple linear regression model that contains multiple independent variables is referred to as
multiple regression model. An equation

Segmentation is the process of segmenting the data according to the company’s requirement
to refine the analyses based on the defined context through certain tools. The Main objective
is to understand the customers better and then obtain actionable data to improve the outcome.
It allows to filter the analyses depending upon some elements.
Object Segmentation is the process of segmenting the moving objects. That means locating
the objects and its boundaries from background. The frames that are captured contain noises
that are extracted from captured frames. The objects after filtering are passed to optical flow
vectors for thresholding operation. The unwanted objects are then removed. In filtering
process, the holes are created. They are closed through a morphological operation.

Dr.G.Naga Satish
Segmentation:
Segmentation is the Process of segmenting the data according to the company’s requirement
to refine the analyses based on the defined context through certain tools.
Objective segmentation is the process of segmenting the moving objects. Locating the objects
and its boundaries from background. The frames that are captured contain noise that are
extracted from captured frames. The Objects after filtering are passed to optical flow vectors
for thresholding operation. The unwanted objects are then removed. In the filtering process
the holes are created. They are closed through morphological operation.

Supervised Learning:

 Supervised learning allows you to collect data or produce a data output from the
previous experience.
 Helps you to optimize performance criteria using experience
 Supervised machine learning helps you to solve various types of real-world
computation problems

Example:

For example, you want to train a machine to help you predict how long it will take you to
drive home from your workplace. Here, you start by creating a set of labeled data. This data
includes

 Weather conditions
 Time of the day
 Holidays
All these details are your inputs. The output is the amount of time it took to drive back home
on that specific day.

Types of Supervised Machine Learning Techniques

Dr.G.Naga Satish
Regression:

Regression technique predicts a single output value using training data.

Example: You can use regression to predict the house price from training data. The input
variables will be locality, size of a house, etc.

Classification:

Classification means to group the output inside a class. If the algorithm tries to label input
into two distinct classes, it is called binary classification. Selecting between more than two
classes is referred to as multiclass classification.

Example: Determining whether or not someone will be a defaulter of the loan.

Strengths: Outputs always have a probabilistic interpretation, and the algorithm can be
regularized to avoid overfitting.

Weaknesses: Logistic regression may underperform when there are multiple or non-linear
decision boundaries. This method is not flexible, so it does not capture more complex
relationships.

Unsupervised Learning

 Unsupervised machine learning finds all kind of unknown patterns in data.

 Unsupervised methods help you to find features which can be useful for
categorization.
 It is taken place in real time, so all the input data to be analyzed and labeled in the
presence of learners.
 It is easier to get unlabeled data from a computer than labeled data, which needs
manual intervention.

Example:
She knows and identifies this dog. A few weeks later a family friend brings along a dog and
tries to play with the baby.
Baby has not seen this dog earlier. But it recognizes many features (2 ears, eyes, walking on 4
legs) are like her pet dog. She identifies a new animal like a dog. This is unsupervised
learning, where you are not taught but you learn from the data (in this case data about a dog.)
Had this been supervised learning, the family friend would have told the baby that it's a dog.

Dr.G.Naga Satish
Unsupervised learning problems further grouped into clustering and association problems.

Clustering

Clustering is an important concept when it comes to unsupervised learning. It mainly deals

with finding a structure or pattern in a collection of uncategorized data. Clustering algorithms
will process your data and find natural clusters (groups) if they exist in the data. You can also
modify how many clusters your algorithms should identify. It allows you to adjust the
granularity of these groups.

Association

Association rules allow you to establish associations amongst data objects inside large
databases. This unsupervised technique is about discovering exciting relationships between
variables in large databases. For example, people that buy a new home most likely to buy
new furniture.

Other Examples:

 A subgroup of cancer patients grouped by their gene expression measurements

 Groups of shopper based on their browsing and purchasing histories
 Movie group by the rating given by movies viewers

Dr.G.Naga Satish
Supervised Learning Unsupervised Learning

Supervised learning algorithms are trained using Unsupervised learning algorithms are
labeled data. trained using unlabeled data.

Supervised learning model takes direct feedback to Unsupervised learning model does not
check if it is predicting correct output or not. take any feedback.

Supervised learning model predicts the output. Unsupervised learning model finds the
hidden patterns in data.

In supervised learning, input data is provided to In unsupervised learning, only input

the model along with the output. data is provided to the model.

The goal of supervised learning is to train the The goal of unsupervised learning is to
model so that it can predict the output when it is find the hidden patterns and useful
given new data. insights from the unknown dataset.

Supervised learning needs supervision to train the Unsupervised learning does not need
model. any supervision to train the model.

Supervised learning can be categorized Unsupervised Learning can be classified

in Classification and Regression problems. in Clustering and Associations problems
.

Supervised learning can be used for Unsupervised learning can be used for those cases
those cases where we know the input as where we have only input data and no
well as corresponding outputs. corresponding output data.

Supervised learning model produces an Unsupervised learning model may give less
accurate result. accurate result as compared to supervised learning.

Dr.G.Naga Satish
Supervised learning is not close to true Unsupervised learning is more close to the true
Artificial intelligence as in this, we first Artificial Intelligence as it learns similarly as a
train the model for each data, and then child learns daily routine things by his
only it can predict the correct output. experiences.

It includes various algorithms such as It includes various algorithms such as Clustering,

Linear Regression, Logistic Regression, KNN, and Apriori algorithm.
Support Vector Machine, Multi-class
Classification, Decision tree, Bayesian
Logic, etc.

Supervised and Unsupervised Learning:

Parameters Supervised Learning Unsupervised Learning

Process Input and Output Variables Only Input Data is Provided
are Provided
Input Data The Algorithms are trained The Algorithms are trained
through labelled data. through unlabelled data.
Number of Classes Number of classes is known Number of classes is not
already known.
Computational Complexity It is a Simple Method It is Complex
Use of Data It uses Training data to learn It does not need any output
a link in between Input and data.
Outputs
Algorithms Used Support Vector Machines, Unsupervised algorithms are
Random Forest, Neural divided into categories like
Network, Classification k-means, cluster algorithms,
Trees, Linear and Logistics hierarchal clsutering
Regression
Real Time Learning Learning methods works Learning methods works in
offline real time
Accuracy of Results It is Accurate and Trusty It is Less accurate and trusty
worthy method worthy method.
Main Drawback Classification of Bigdata is a Because data used in

Dr.G.Naga Satish
 challenging task unsupervised learning is
labelled and unknown it is
difficult to obtain precise
information regarding the
data sorting.

TREE BUILDING
Decision Tree is a diagrammatic representation of alternative courses of action and sequence
of states of nature.
The Various steps involved in decision tree analysis are
1. Determine the number of decisions to be taken and the alternative strategies available
for each decision in sequential manner.
2. Determine the outcome( or event) which may occur from each alternative strategy.
3. Construct a tree diagram representing the order in which decisions are taken and
outcomes are occurring. The decision tree diagram begins from left side and move
towards right side.
4. Determine the Probabilities of occurrences of each state of nature.
5. Determine the pay off values for each pair ( or combination) of state of nature and
course of action.
6. Calculate expected pay off value for each course of action starting from right side of
the decision trees.
7. Select the course of action (or alternative strategy) with the best expected pay off
value.
8. With backwards from last decision point to first decision point and at each decision
point repeat the steps from step 4 to step7.

Dr.G.Naga Satish
Time Series Methods:
The time series data is ordered sequence of observations on quantative variable that is
measured over equally spaced time interval. The time series are used in statistics, signal
Processing, econometrics, pattern recognition, weather forecasting.
The time series analysis is used to analyse the time series data and to forecast the future value
of variable under consideration. The data in time series analysis contain set of identifiable
components and random errors that make the pattern difficult to identify.
ARIMA is one of the time series methods. ARIMA stands for Autoregressive Integrated
Moving Average. In time series analyses it is generalization. This model is fitted to time
series data to understand the data or to predict future points in the series. They are applied in
certain cases where the data shows evidence of non stationary and where the initial
differencing step is applied to minimize the non stationary.
ARIMA is well known stochastic method that is used to analyze the time series data sets. It
contains three time series components such AR (Auto-Regressive), I (Integrated), MA
(Moving Average). Every component minimizes the final residuals indicated as p, d, q
respectively. Integrated (I) is initial step in ARIMA to extract trend data. This is possible by
differencing the data from earlier values. The first step is indicated by (0,1,0) and the second
step is indicated by (0,2,0). This continues until data becomes trend less. The time series will
become trend less after differencing. The second step in ARIMA is auto regression to analyze
the time series data. Once this data becomes stationary the AR component gets activated. The
auto regressive step will extract the previous value from present value. This is obtained
through simple linear regression by considering independent or predictor variables as it time
lagged values.

Y=C+Y+Y+Y…..+,ib

Dr.G.Naga Satish
OVERFITTING
The Decision Trees will be at high risk to overfit the training data to a high degree when they
are not pruned.
Over fitting might occurs when models select the noise or errors in training data set.
Therefore overfitting will be seen viewed as performance gap between training and test data.
A General method to reduce overfitting in decision trees is decision tree pruning. There are
two methods namely
1. Post Pruning
2. Pre Pruning

The above figure shows relationship between Tree Depth and Overfitting

The following figure shows the Reduced Error Pruning

Dr.G.Naga Satish
STL APPROACH:
STL Stands for Seasonal and Trend Decomposition using Loess. Loess is a method used to
estimate the non linear relationships. STL is versatile, robust ad stastical method of
decomposing the time series data into three components as Trend, Seasonality and residual.
The purpose of development of STL is to build a decomposition procedure and a computer
implementation that can satisfy the below criteria.
1. A Simple design and a straight forward use.
2. Easy computer implementation and a fast computation for long time series.
3. Specification of observations for each cycle of seasonal component to an integer greater
than 1.
4. Flexible to specify the variation in trend and seasonal components.
5. Robust trend and seasonal components that cannot be distorted by transient and aberrant
behaviour of data.
6. It should be able to decompose the series with missing values.
STL contains a set of smoothing operations that employ same smoother locally weighted
regression or loess with only one exception.
STL has various parameters that must be selected by the data analysts.

n(p)= Number of observations in every cycle of seasonal component.

n(i)= Number of Passes through inner loop.
n(o)= Number of robustness iterations of outer loop.
n(l)= Smoothing parameters for low pass filter
n(t)= Smoothing parameter for trend component.
n(x)= Smoothing parameter for seasonal component.

STL can be implemented in any environments for the purpose of graphics and data analysis.

Dr.G.Naga Satish
Pruning:
Pruning is a technique that is used to reduce the size of decision tress by removing parts of
the tree which provide less power for classifying instances. It reduces the complexity of final
classifier and even improves the accuracy by reduction of over fitting.

Pruning occurs either in top-down or bottom-up. The top-down pruning traverses the nodes
and trims the sub tree starting at root. The bottom up pruning begins at leaf nodes.

Pruning when applied on decision trees will remove one or more sub trees from it. There are
various methods for decision tree pruning .They replaces the sub tree with leaf if
classification accuracy is not reduced over pruning data set. Pruning increases number of
classification errors on training set but improves classification accuracy on unseen data The
Pruning techniques can be divided into two groups. The methods in first group will compute
the probability of sub tree misclassification and then make the pruning decision through an
independent test set called pruning data set. In Second group the iterative grow and prune
method is used while creating a tree. These pruning techniques are as follows.

1. Cost Complexity Pruning

2. Reduced Error Pruning
3. Critical Value Pruning
4. Minimum Error Pruning
5. Pessimistic Error Pruning
6. Error Based Pruning
7. Optimal Pruning
8. Minimum Description Length Pruning

Dr.G.Naga Satish
Extract features from generated model as Height, Average Energy etc and Analyze for
prediction

Feature extraction is the process of extracting useful characteristics from data. It calculates
the values from input images. A feature is also called as descriptor. It is defined as a function
of one or more measurements by specifying certain quantifiable property of complete image
or sub image or of single object. Three methods of feature extraction are performed that lead
to three different results for every classification. They are compared each other. The best
feature extraction method for very two classifications is used for the part and complete
system together.
For example consider an image model that is generated. The feature extraction focuses
around the measurement of geometric properties and surface characteristics of regions. The
features extracted from this image model are

Type Feature Description

Y ( i, j) Center of Gravity
I, R ,G, B (G) Mean Gray Value
L L*a*b Color Components of region
Y h,w, A, L,R Height, width, area,
roundness, perimeter
Y a Average
Y E Energy
I, R, G,B K Deviation Contrasts

The features provide relevant information for classification. To reduce the computational
time required, in pattern recognition process it is necessary to select features suitable for
classification.
The extracted features are later on analysed for prediction. A Predictor is used for this
purpose. The model is trained on labelled examples can be good and bad quality to retrieve
quality classification model. A model is trained on labelled examples of salient and non-
salient images for retrieving content classification model. While using a model to classify the
data the output of image is a class label and a score matrix.
The predictor is selected due to its advantages. The advantage does not have the problem of
Overfitting. The Overfitting occurs when model has multiple features related to the number
of observations and results in poor predictive performance. This problem is relevant to
consider while working with machine learning on images because number of features
extracted from an image are large.

Dr.G.Naga Satish
Measures of Forecast Accuracy:
Forecast accuracy is a method of deviation of prediction or forecast from the actual outcome.
It is generally defined as
Error=Actual Demand-Forecast
Or
e=A-F

Forecast accuracy can be measured by using two methods

1. Mean Forecast Accuracy(MFE)
2. Mean Absolute Deviation (MAD)

Mean Forecast Accuracy (MFE):

Mean forecast error represents the deviation of forecast from actual demand. It is the mean of
differences for every period in between number of period forecasts and actual demand for
related periods. This error is mostly used as bias for following and adjusting forecasts. If it is
positive then the forecasts are low compared to actual demand. And if it is negative then the
forecasts will be high. Mean forecast deviation is defined as follows
∑ ( )
MFE=

If the value of MFE is be greater than zero then model tends to under forecast and if the value
of MFE is less than zero then model tends to over forecast.

Mean Absolute Deviation (MAD)

Mean absolute deviation deviates the forecasted demand from actual demand. It is the mean
deviation for every period absolute terms and respective period demand.

MAD=∑ |𝑒𝑖 |

Uses of Forecast Error:

1. It is used to compare alternative forecasting models

2. It is used to forecast the model bias.
3. It is used to generate absolute size of forecast errors.
4. It is used to identify the forecast models which needed to be adjusted.

Dr.G.Naga Satish