An Improved Intrusion Detection in Wireless Sensor Networks Using Hybrid Multiclass Over-Sampling and Deep Neural Networks

Abstract

With the emergence of new attacks, there is a continual need for innovative approaches that can closely monitor and swiftly adapt to evolving threats. IDSs can be broadly categorized into misuse detection and anomaly detection, each utilizing machine-learning methods. Machine learning algorithms, particularly those relying on datasets like DARPA and KDD Cup 1999, have gained popularity. However, challenges include dataset limitations, overfitting, and the requirement for substantial computational power. This study focuses on the specific problem of class imbalance in Wireless Sensor Networks (WSN) datasets for intrusion detection. Existing techniques, such as oversampling and undersampling, have limitations, and the imbalance poses challenges for accurate intrusion detection model development. The research aims to develop an enhanced intrusion detection model addressing multiclass imbalance through an optimized KNN-SMOTE oversampling technique integrated with a Deep Learning model. The significance lies in its potential to greatly enhance the accuracy of intrusion detection systems, contributing to the security of computer networks. The study's scope involves static dataset analysis and multiclass classification of intrusion attacks. The research questions revolve around the development of a multiclass oversampling technique, a hybrid model, and the performance evaluation of the developed system in comparison to existing IDS. The research proposes a novel OK-SMOTE-DL model combining the potential of the KNN model, firefly algorithm, and the SMOTE model fused with Deep Learning techniques to help create a better model for handling multiclass intrusion datasets that can similarly detect minority classes effectively in wireless sensor networks.