Performance Analysis of Machine Learning Algorithms for Security and Privacy Vulnerabilities in IoT

Abstract

The rapid growth of the Internet of Things (IoT) has brought forth numerous security and privacy challenges. In this MS thesis, we propose a comprehensive research study on the performance analysis of machine learning algorithms for addressing these vulnerabilities in the IoT environment. We employ three datasets, namely the NSL-KDD Dataset, CIC_IoT_2022 Dataset, and a SQL injection dataset, to evaluate the effectiveness of various machine learning algorithms. To ensure the reliability of our analysis, the datasets undergo multiple pre-processing operations, including data cleaning, normalization, and feature engineering. Dominant features are then extracted to develop benchmark datasets that capture the essential characteristics of IoT security and privacy vulnerabilities. Multiple classification algorithms, including Decision Tree, Support Vector Machines (SVM), K-Nearest Neighbors (KNN), and Logistic Regression, are applied to the benchmark datasets. The performance of these algorithms is evaluated using standard evaluation metrics such as accuracy, precision, recall, F1-score, and area under the curve (AUC). In addition, we explore NLP-based algorithms specifically designed for the SQL injection dataset to detect and mitigate SQL injection attacks. The dataset was extensively analyzed using the python Plotly library where histograms,Using feature extraction, the major vulnerabilities were extracted as specific attributes later used to train the machine learning model separately. In conclusion, the proposed methodology demonstrates enormous potential for practical applications in improving the security and privacy of IoT systems. By leveraging machine learning algorithms and conducting a comprehensive analysis, we can effectively detect and mitigate security vulnerabilities, safeguarding IoT devices and networks. The insights gained from this research contribute to the development of robust security mechanisms and guide the implementation of proactive measures to protect IoT systems from potential attacks. Future work includes exploring advanced machine learning techniques such as deep learning and ensemble methods to further enhance the performance of algorithms in addressing IoT vulnerabilities. Additionally, investigating the impact of different feature engineering techniques and exploring real-time detection and response mechanisms can lead to more effective security measures. Overall, this research paves the way for future studies on securing IoT systems and promotes the adoption of machine learning in IoT security and privacy. The applied algorithms perform well for the datasets, where Decision Tree algorithm surpassed the other algorithms for both NSL-KDD and CIC-IoT2022. For NSL-KDD dataset Decision Tree produced a training and validation accuracy of 99.00% and 98.54% respectively. For CIC-IoT2022 dataset Decision Tree produced a training and validation accuracy of 98.92% and 98.55% respectively. For the SQL Injection Attack dataset Naïve Bayes produced the best accuracy of 99.36% for training process and 98.13% for validation process, respectively.