Leveraging Machine Learning for Behavioral Analysis and Mitigation of APT Attacks in WSNs

Abstract

Advanced Persistent Threats (APTs) pose a significant challenge to cybersecurity especially in Wireless Sensor Networks (WSNs) due to their sophisticated, prolonged and targeted nature. Traditional detection methods often struggle to identify and counteract these evasive threats effectively. This paper explores the application of ML (Machine Learning) models for behavioral analysis of APT attacks, aiming to enhance detection and response mechanisms through the enhanced dataset using CGAN (Conditional Generative Adversarial Network) models. It mainly focuses on integration of the dataset generated using CGAN with ML classification techniques. Also on exploration of SCVIC-APT-2021 dataset along with all its features based on CICFlowmeter-V4.0 and applying classification techniques such as SVM (Support Vector Machine), RF(Random Forest) and KNN(K Nearest Neighbor) to label arrived sample from the environment. The proposed model has made a comparative analysis corresponding to performance on precision and f1-scores of the classification models. The paper also evaluates feature selection techniques and data preprocessing methods to improve model accuracy and reduce false positives. Experimental results demonstrate that ML-based approaches can significantly outperform traditional signature-based methods in detecting and mitigating APTs, offering more adaptive and scalable solutions. The findings highlight the importance of continuous learning and model updating to keep pace with evolving APT tactics. This paper contributes to the growing system of knowledge in cybersecurity by providing insights into the practical implementation of machine learning models for Advanced Persistent Threat detection and presenting a framework for integrating these models into existing security infrastructures. The practical results indicate that among SVM, RF and KNN, RF has emerged as the most suitable classifier for derived datasets using CGAN models.