ENSEMBLE-LEARNING-BASED DEEP NEURAL NETWORK ATTACK CLASSIFICATION OF IMBALANCED IOT INTRUSION DATA

Abstract

IoT gadgets feature the requirement for solid safety efforts to lessen weaknesses and risks in connected networks. This paper presents a Bagging Classifier (BC)-based Deep Neural Network (DNN) strategy to deal with class irregularity worries in IoT intrusion detection datasets. This strategy utilizes deep learning and ensemble learning to improve intrusion detection and arrangement. Four unmistakable ID datasets — NSL-KDD, KDDCUP99, UNSW-NB15, and Bot-Io — show promising accuracy, precision, recall, Fscore, and false positive rate. The recommended technique beats past strategies, particularly while involving 10 base assessors in the bagging ensemble approach. Further exploration utilizes Convolutional Neural Networks (CNN) and hybrid CNN + Long Short-Term Memory (LSTM) models to accomplish close to 100% accuracy. Flask is utilized to give a front-end connection point to user testing and verification, further developing convenience. This exploration further develops IoT ID by showing ensemble learning's capacity to deal with class unevenness issues and further develop network security.