IoT Security Enhancement Using Machine Learning Algorithms

Research Paper Summary

Karthik H
(VAS22CS074)
Guided By: Asst. Prof Anitha A S

S7 CSE B
Vidya Academy of Science and Technology, Thalakotukkara

September 10, 2025

Presentation Outline

Introduction

Literature Review

Methodology

Results

Conclusion

References
Research Problem

• IoT devices face increasing security challenges across various sectors

• Smart cities, homes, healthcare, and transportation

• Traditional security measures are inadequate for autonomous IoT devices

• Growing reliance on cloud computing and SDN leads to more security threats

• Most IoT devices operate without human intervention, increasing vulnerability

Introduction 3/22
Types of Security Attacks

Figure: Types of possible security attacks

Introduction 4/22
Research Objectives

• Present a novel model for enhancing IoT security using ML classifiers

• Analyze recent technologies, security challenges, and vulnerabilities in IoT systems

• Develop a cyberattack detection solution for IoT devices using ML

• Identify the most accurate ML classifiers for AI-based reaction agents

• Create autonomous security management systems for IoT networks

Introduction 5/22
Significance of the Research

• Highlights benefits and limitations of applying ML in IoT environments

• Proposes ML-based security model that autonomously manages security issues

• Improves effectiveness of cybersecurity systems and IoT infrastructure

• Demonstrates superior performance in execution speed and accuracy

• Provides foundation for next-generation IoT security systems

Introduction 6/22
ML Techniques for IoT Security

• Intrusion detection and prevention: Network traffic and device log analysis

• Anomaly detection: Real-time data analysis for abnormal patterns

• Threat intelligence and prediction: Big security dataset analysis

• Firmware and software vulnerability analysis: Identifying security flaws

• Behavior-based authentication: Analyzing device usage patterns

• Data privacy and encryption: Secure transmission and storage

Literature Review 7/22

Existing Models and Approaches

• LSTM, CNN, and hybrid approaches for identifying network breaches in IIoT

• LEGATO algorithm for reducing user dropout in federated learning

• Auditable privacy-preserving federated learning (AP2FL) for medical devices

• SteelEye semi-deep learning approach for industrial control systems

• Fuzzy Deep Learning models for network risk detection

Literature Review 8/22

IoT Vulnerabilities and Attacks
• Vulnerabilities:
• Unpatched vulnerabilities in outdated software
• Weak authentication mechanisms
• Susceptible Application Programming Interfaces (APIs)

• Attack Types:
• Botnets (e.g., Mirai)
• Sleep deprivation attacks
• Node tampering and jamming
• Eavesdropping

• Layer-specific Attacks:
• Network layer attacks targeting routers and bridges
• Application layer attacks (malware, phishing, code injection)

Literature Review 9/22

Machine Learning Approach

• Evaluation of multiple ML classifiers for cyberattack detection accuracy

• Implementation of AI-based reaction agents for pattern recognition

• Development of autonomous security management systems

• Integration of ML techniques with IoT security infrastructure

• Multi-layer security assessment across IoT architecture

Methodology 10/22
Data and Analysis Framework

• Dataset: IoT Dataset for Intrusion Detection Systems (IDS) from Kaggle

• Performance Metrics: Execution speed and accuracy comparisons

• Multi-layer Analysis: Security assessment across different IoT layers
• Physical, data, network, cloud, application layers

• Integration Approach: Comprehensive defense system with capabilities for:

• Real-time anomalous behavior detection
• Security threat prediction
• Automatic response to identified attacks
• Continuous learning and adaptation

Methodology 11/22
Proposed Security Framework

Methodology 12/22
Performance Evaluation Framework
• Evaluation Metrics:
• Accuracy, Precision, Recall, F1-score
• AUC (Area Under ROC Curve)
• Execution Time, Cost Per Example (CPE)

• Datasets Used:
• UNSW-NB15: 49 features, 1.5M+ instances
• BoTNet-IoT-L01: 23 features, 1.1M+ instances
• NSL-KDD: 42 features, 148K+ instances

• Model Configuration:
• RF-BPNN: 500 epochs, 1-4 neural network layers
• SMOTE for class imbalance handling
• Cost Matrix Analysis for evaluation

Results 13/22
Overall Performance Results

Metric Random Forest XGBoost AdaBoost Ensemble RF-BPNN T

Accuracy 99.9% 99.9% 99.8% 99.9%

Detection Rate 99.8% 99.8% 99.7% 99.8%
F1-score 99.9% 99.9% 99.8% 99.9%
AUC 1.0 1.0 0.99 1.0
False Positive Rate 0.1% 0.1% 0.2% 0.1%

Table: Overall performance comparison of ML models

Results 14/22
Dataset-Specific Performance

Dataset Best Model Accuracy Key Strength

UNSW-NB15 Random Forest 99.9% Excellent across all attack types

BoTNet-IoT-L01 Ensemble RF-BPNN 99.2% High sensitivity (97.6%)
NSL-KDD Random Forest 99.8% Best for U2L attacks (99.4%)

Table: Performance across different datasets

Results 15/22
Attack-Type Specific Performance

Attack Type Random Forest J48 Bayes Net Hoeffding Tree

U2L 99.4% 99.3% 97.2% 97.8%

U2R 83.4% 82.1% 75.6% 79.3%
Probe 99.8% 99.7% 98.2% 98.9%
DoS 100% 99.9% 99.1% 99.5%
Normal 99.6% 99.5% 98.3% 98.7%

Table: Precision by attack type on NSL-KDD dataset

Results 16/22
Computational Efficiency

Algorithm CPE (%) Time (seconds)

Random Forest 0.2% 73.8

J48 0.5% 43.6
Bayes Net 6.8% 5.7
Hoeffding Tree 7.5% 4.6

Table: Computational performance metrics

Results 17/22
Performance Improvements Summary

• 99.9% Accuracy: Exceptional classification performance

• Perfect AUC Score: Optimal discrimination capability

• Low False Positive Rate: Minimized false alarms (0.1%)

• Comprehensive Attack Detection: High accuracy across diverse attack types

• Computational Efficiency: Balanced performance-speed trade-offs

• Scalability: Consistent performance across large-scale datasets

Results 18/22
Main Conclusions

• ML Integration: Machine learning significantly enhances IoT security when

properly integrated
• Autonomous Management: ML-based models effectively manage security
challenges without constant human oversight
• Performance Superiority: The proposed approach outperforms existing
ML-based security models
• Comprehensive Protection: Multi-layer security analysis provides better overall
protection
• Scalability: Autonomous nature makes it suitable for large-scale IoT deployments

Conclusion 19/22
Implications and Future Work
• Practical Applications: Applicable across healthcare, smart cities, and industrial
systems
• Industry Impact: Contributes to more secure and reliable IoT systems for
widespread adoption
• Limitations:
• Need for continuous model updates against emerging threats
• Integration challenges with existing IoT infrastructure
• Balancing security measures with system performance

• Future Research Directions:

• Adaptive learning mechanisms for evolving threats
• Lightweight security solutions for resource-constrained devices
• Integration with blockchain for enhanced security

Conclusion 20/22
Key References
Abbas, Y., Ali, D. & Gautam, S. AP2FL: Auditable privacy-preserving federated learning framework for
electronics in healthcare. IEEE Trans. Consumer Electron. 99, 1 (2023).
Danyal, N., Abbas, Y., Ali, D. & Gautam, S. Federated quantum-based privacy-preserving threat detection
model for consumer Internet of Things. IEEE Trans. Consumer Electron. (2024).
Özalp, A. N. et al. Layer-based examination of cyber-attacks in IoT. In 2022 International Congress on
Human-Computer Interaction, Optimization and Robotic Applications (HORA) (IEEE, 2022).
Altunay, H. C. & Albayrak, Z. A hybrid CNN+ LSTM-Based intrusion detection system for industrial IoT
networks. Eng. Sci. Technol. Int. J. 38, 101322 (2023).
Sanaz, N., Behrouz, Z., Abbas, Y. & Ali, D. Steeleye: An application-layer attack detection and
attribution model in industrial control systems using semi-deep learning. In 2021 18th International
Conference on Privacy, Security and Trust (PST), IEEE Xplore (2021).
Abbas, Y., Ali, D., Reza, M. P., Gautam, S. & Hadis, K. Secure intelligent fuzzy blockchain framework:
Effective threat detection in IoT networks. Comput. Ind. 144, 103801 (2023).
Otoum, Y. & Nayak, A. On securing IoT from deep learning perspective. In Proc. 2020 IEEE Symposium
on Computers and Communications (ISCC) 1-7 (2020).
IoT Dataset for Intrusion Detection Systems (IDS). Available at:
https://www.kaggle.com/azalhowaide/iot-dataset-for-intrusion-detection-systems-ids (2023).
References 21/22
Thank You
Questions?