DEPARTMENT OF ELECTRONICS AND COMMUNICATION

ENGINEERING

Project phase-II

URL-DRIVEN DETECTION OF MALICIOUS

BEHAVIOUR IN SOCIAL NETWORK

PRESENTED BY GUIDED BY
RAJESWARI E [20UEC137] PRIYA R
MONISHA M [20UEC094] Assistant Professor
SINDHU E [20UEC159]

04/30/2025 1
 OUTLINE OF THE PRESENTATION

• Introduction
• Literature survey
• Problem Definition
• Objective
• Block Diagram
• Input Specification
• Implementation
• Result Inferences
• Graph
• Comparison Graph
• Conclusion
• Reference
04/30/2025 2
 INTRODUCTION

• Malicious URLs present serious risks in the world of digital networks since
they act as trickery access points for fraud, cyberattacks, and scams.
• Malicious social bot is a software program that pretends to be a real user in
online social networks (OSNs) .
• Moreover, malicious social bots perform several malicious attacks, such as
spread social spam content, generate fake identities, manipulate online ratings,
and perform phishing attacks.
• when a user wants to share a tweet containing URL(s) with the neighbouring
user, the participant adapts URL shortened service (i.e., bit.ly in order to reduce
the length of URL (because a tweet is restricted up to 140 characters).

04/30/2025 3
 LITERATURE SURVEY
SI.NO TITLE AUTHOR YEAR ALGORITHM ADVANTAGE DISADVANTAGE
USED
1. Malicious URL T Shantanu, Janet 2023 Random Forest Through the comparison of Traditional techniques like
Detection: A B different approaches, the blacklisting, regular
Comparative system helps identify expressions, and signature
Study methods that offer higher matching struggle due to
accuracy and efficiency in evolving patterns, large data
detecting malicious URLs. volumes, and complex
feature relationships in
detecting malicious URLs.

2. Malicious URL Cho Do Xuan1 , 2023 Support Machine learning models can The effectiveness of machine
Detection based Hoa Dinh Nguyen machine vector scale to handle large volumes learning models heavily
on Machine of URLs, making them relies on the quality and
Learning suitable for both small-scale representativeness of the
and large-scale applications. training data

3. Hybrid Solution A. Mishr and B. 2022 Link Guard They can incorporate new Hybrid solutions often
to Detect and B. Gupta algorithm detection techniques and involve multiple components
Filter Zero-day update their algorithms to and layers of security
Phishing stay effective against measures, making them
Attacks evolving attack strategies. inherently more complex to
design, implement, and
maintain.

04/30/2025 4
 LITERATURE SURVEY
SI.NO TITLE AUTHOR YEAR ALGORITHM ADVANTAGE DISADVANTAGE
USED

4 A Hybrid Approach Namrata Singh1 , 2022 Link Guard This reduces the risk of This complexity can make
to Detect Zero Day algorithm false negatives and the system harder to
Phishing Websites Nihar Ranjan Roy improves overall manage, maintain, and
security. troubleshoot.

5 Phish catcher: Client- Muzammil Ahmed, 2021 Naive Bayes By utilizing machine Overfitting occurs when
side Defence Against Ahme Altamimi, classifiers learning algorithms, a model performs
Web Spoofing Wilayat Khan, Phish Catcher can adapt exceptionally well on the
Attacks Using Mohammad and evolve to recognize training data but struggles
Machine Learning Alsaffar new and evolving web with new, unseen data.
spoofing techniques.
6 Malicious URL U. S. D. R, A. 2021 Gradient Boosting Automated detection Limited efficiency against
Detection and Patil , Mohana Classifier saves time.ML models novel attacks. Vulnerable
Classification adapt to new to adversarial inputs. High
Analysis using threats .Increased the false positive rates in
Machine Learning accuracy reduces false complex environments.
Models positives. Scalable for
large datasets

7 2021 Unsupervised its simplicity and ease of It assumes a linear

Malicious URL A. Lakshmanarao,
Learning algorithm implementation, making relationship between
Detection using NLP, M. R. Babu the computationally features and the log-odds
efficient for large of the target variable,
Machine Learning
datasets. which may not hold true
and FLASK for complex, nonlinear
relationships in URL data.

04/30/2025 5
 LITERATURE SURVEY
SI.NO TITLE AUTHOR YEAR ALGORITHM ADVANTAGE DISADVANTAGE
USED

8 Malicious URL R. Chiramdasu, G. 2021 Naive Bayes • Scalability This can lead to lower
Detection using Srivastava Classifier • Robustness accuracy when the data is
Logistic • Efficiency highly complex.
Regression

9 Intelligent Y. -C. Chen, Y. -W. 2020 Link Guard logistic regression can Hybrid solutions often
Malicious URL Ma algorithm handle both binary and involve multiple
Detection with multiclass classification components and layers of
Feature Analysis tasks, providing security measures, making
flexibility in addressing them inherently more
different threat scenarios complex to design,
implement, and maintain.
10 Malicious URL: M. Mehndiratta, N. 2020 Feature Selection Its requires relatively low its inability to capture
Analysis and J and Dimensionality training time and complex nonlinear
Detection using Reduction resources compared to relationships between
deep learning more complex models, features and labels, which
making it suitable for can lead to lower accuracy.
real-time or large-scale
applications.

04/30/2025 6
 PROBLEM DEFINITION

• Complex models like gradient boosting and the hybrid LSD model can
get too focused on the training data, kind of like memorizing answers
instead of understanding concepts.
• This can be a problem because when faced with new, unfamiliar data
(like new types of phishing attacks), these models might not know what
to do since they haven't learned to adapt.

• The malicious bots generate fake news and creating multiple fake
account.

• No privacy and the security are less.

04/30/2025 7
 OBJECTIVE OF THE PROJECT

To create a strong cybersecurity solution by using advanced neural

network techniques that learn from both past and future data, helping to
detect and defend against evolving phishing threats more effectively.

04/30/2025 8
 BLOCK DIAGARM

04/30/2025 9
 INPUT SPECIFICATIONS

• SOCIAL MEDIA DATA: The system will process data from various
social media platforms (e.g., Twitter, Facebook, Instagram). This may
include user profiles, posts, comments, likes, and network relationships.
• URLs: Extracted URLs from social media posts and profiles.

04/30/2025 10
 IMPLEMENTATION STEPS

04/30/2025 11
 BILSTM ALGORITHM

The key feature of a BiLSTM is that it consists of two LSTM layers: one
processing the input sequence a forward direction, and the other processing it in a
backward direction. The forward LSTM processes the input sequence from the
beginning to the end, while the backward LSTM processes it in the reverse order,
starting from the end and moving towards the beginning.

04/30/2025 12
 GRAPH
ACCURACY GRAPH OF
HYBRID MODEL

04/30/2025 13
 FORMULA USED

• Accuracy: It measures the overall correctness of the model and is

calculated as the ratio of correctly predicted instances to the total instances.

𝑻𝑷+ 886 + 1186

𝑻𝑵
=
𝑻𝑷+𝑻𝑵+𝑭𝑷 886 + 90 + 49 +
Accuracy = 0.9371
= +𝑭𝑵 1186

• Precision: It measures the accuracy of positive predictions

and is calculated as the ratio of true positives to the sum of
true positives and false positives.
𝑻 88
Precision = 𝑷
𝑻𝑷+ 6
886+9
=
𝑭𝑷 0
0.9077

04/30/2025
 • Recall (Sensitivity): It measures the proportion of actual
positive instances that are correctly predicted and is
calculated as the ratio of true positives to the sum of true
𝑻𝑷
positives and false negatives.886
𝑻𝑷+𝑭𝑵 886+49 = 0.9475
Recall =

• F1 Score: It is the harmonic mean of precision and recall

and provides a balance between the two. It is calculated as:

𝑷𝒓𝒆𝒄𝒊𝒔𝒊𝒐𝒏 x 𝑹𝒆𝒄𝒂𝒍𝒍 = 2 x 𝟎.𝟗077 x = 0.9271

𝟎.9475
F1-Score = 2 x

• CONFUSION MATRIX: 𝑷𝒓𝒆𝒄𝒊𝒔𝒊𝒐𝒏 + 𝑹𝒆𝒄𝒂𝒍𝒍

True 𝟎.𝟗077
Positive+𝟎.9475
(TP): Predicted positive and positive
True Negative (TN): Predicted negative and negative
False Positive (FP): Predicted positive but negative
False Negative (FN): Predicted negative but positive

04/30/2025
COMPARISON OF EXISTING AND
PROPOSED WORK

EVALUATION METRIC VALUE

Existing Work Proposed Work

Metric (%) (%)
Accuracy 88 94
Precision 84 90
F1-Score 83 94
Recall 85 92
 COMPARISON GRAPH

04/30/2025 17
 RESULT INFERENCES

04/30/2025 18
 CONCLUSION
• The proposed cybersecurity system, integrating BiGRU and
BiLSTM networks via transfer learning, greatly boosts accuracy and
precision, offering robust defense against evolving cyber threats by
detecting and mitigating phishing URLs and malicious links.
• Its adaptability and continuous learning ensure resilience in dynamic
environments, validated by comprehensive performance analysis,
making it a pivotal solution for enhancing cybersecurity and
combatting cyber threats effectively in an interconnected world.

04/30/2025 19
 REFERENCES

[1]. R. R. Rout, G. Lingam and D. V. L. N. Somayajulu, "Detection of Malicious Social Bots

Using Learning Automata With URL Features in Twitter Network," in IEEE Transactions
on Computational Social Systems, vol. 7, no. 4, pp. 1004-1018, Aug. 2022.
[2]. M. Aljabri et al., "Detecting Malicious URLs Using Machine Learning Techniques:
Review and Research Directions," in IEEE Access, vol. 10, pp. 121395-121417, 2022.
[3]. X. Yan, Y. Xu, B. Cui, S. Zhang, T. Guo and C. Li, "Learning URL Embedding for
Malicious Website Detection," in IEEE Transactions on Industrial Informatics, vol. 16,
no. 10, pp. 6673-6681, Oct. 2020.
[4]. J. Yuan, G. Chen, S. Tian and X. Pei, "Malicious URL Detection Based on a Parallel
Neural Joint Model," in IEEE Access, vol. 9, pp. 9464-9472, 2021.
[5]. S. Asiri, Y. Xiao, S. Alzahrani, S. Li and T. Li, "A Survey of Intelligent Detection Designs
of HTML URL Phishing Attacks," in IEEE Access, vol. 11, pp. 6421-6443, 2023.

04/30/2025 20
 REFERENCES

[6]. A. A. Rahmanian, M. Ghobaei-Arani, and S. Tofighy, ―A learning automata-based

ensemble resource usage prediction algorithm for cloud computing environment,
Future Gener. Comput. Syst., vol. 79, pp. 54–71, Feb. 2022.
[7]. A. Moayedikia, K.-L. Ong, Y. L. Boo, and W. G. S. Yeoh, ―Task
assignment in microtask crowdsourcing platforms using learning automata, Eng.
Appl. Artif. Intell., vol. 74, pp. 212–225, Sep. 2021.
[8]. G. Lingam, R. R. Rout, and D. Somayajulu, ―Learning automata based trust model
for user recommendations in online social networks, Comput. Electr. Eng., vol. 66, pp.
174–188, Feb. 2020.
[9]. Manju, S. Chand, and B. Kumar, ―Target coverage heuristic based on learning
automata in wireless sensor networks, IET Wireless Sensor Syst., vol. 8, no. 3, pp.
109–115, Jun. 2018.
[10]. N. Abokhodair, D. Yoo, and D. W. McDonald, ―Dissecting a social botnet:
Growth, content and influence in Twitter, in Proc. 18th ACM Conf. Comput.
Supported Cooperat. Work Social CompuSt., 2018, pp. 839–851.

04/30/2025 21