SWARM BASED FEATURE SELECTION AND ENSEMBLE DEEP LEARNING MODEL (EDLM) FOR BOTNET ATTACK DETECTION IN IoT HEALTHCARE SYSTEMS

Abstract

A variety of devices make up the Internet of Things (IoT). Because of the large number of attack routes and the constant growth of viruses, botnet detection (BND) is getting more and more difficult. Because of the complexity and diversity of attacks, conventional detection methods that rely solely on a single Machine Learning (ML) technique could not be completely effective. The Ensemble (DL) Deep Learning Model (EDLM) is presented in this study. Divergence Weight (LSTM) Long Short-Term Memory (DWLSTM), Levy Weight (LW) Bi-Directional Gated Recurrent Unit (LWBi-GRU), and CDBN are some of the models that are combined to produce the findings of this EDL. To choose the most relevant features from the dataset, the Inertia Weight Mother Optimisation Algorithm (IWMOA) is presented. The data sequences are processed in both forward (F) and backward (B) directions by the DWLSTM classifier. An update gate (UG) and a reset gate (RG) process LWBi-GRU in both directions (F and B). This classifier's capacity to analyse information from both sides allows it to interpret some inputs effectively. The Conditional Gaussian-Bernoulli Restricted Boltzmann Machine (CGBRBM) for botnet attack detection (BN AD) is a component of the Conditional Deep Belief Network (CDBN). Here, the ensemble averaging (EA) method (EAM) is utilized for the purpose of integrating benefit of many classifiers. The diversity of samples from multiple sources are detected by the potential of the ensemble model (EM) via averaging the predictions of various models. From the Kaggle and the University of California, Irvine (UCI), the Bot-IoT and N-BaIoT datasets are used, and it may simulate attack scenarios. In the IoT datasets, the testing data was used to assess the DL models, and the metrics like precision (P), recall (R), F1-score, and accuracy (ACC) were used for this evaluation.