Machine Learning Techniques for Enhanced Intrusion Detection in IoT Security
Hanadi Hakami, Muhammad Faheem, Majid Bashir Ahmad
Abstract
Network Intrusion Detection Systems (NIDSs) are fundamental to safeguarding computer networks. Intrusion detection systems must become more effective as new attacks are developed and networks grow. Anomaly-based automated detection stands out due to its superior performance among the various detection techniques. However, with the increasing complexity and frequency of cyberattacks, managing vast amounts of data remains challenging for anomaly-based NIDS. Therefore, it is necessary to find an efficient method for solving the problem by using classification with an intrusion detection system which analyzes enormous amounts of traffic data. This research introduces a new model that leverages machine learning (ML) and deep learning (DL) to enhance detection effectiveness and ensure reliability. The approach optimizes data preprocessing by integrating SMOTE for effective data balancing and Pearson’s Correlation Coefficient (PCC) for feature selection. We compared several ML and DL techniques to detect and address the most efficient one for our pipeline. Compared with other approaches, LSTM and RF show superior results when tested on the WSN-DS, UNSW-NB15, and CIC-IDS 2017 datasets. Additionally, the proposed solution prevents biases from arising by addressing imbalanced datasets.