Deep Learning Approach for Protecting IoT Smart Home Devices Against Multiclass Botnet Attacks

Abstract

As the Internet of Things (IoT) continues to expand, smart homes have emerged as a prominent application area, characterized by a network of interconnected devices designed to improve daily life through automation, enhanced security, and energy management. Despite these benefits, the integration of numerous connected devices also brings about significant security challenges, particularly the threat of botnet attacks capable of disrupting entire device networks. The growing number and variety of smart home devices make it increasingly complex to establish effective security mechanisms. To address this concern, this study investigates the use of deep learning (DL) models in combination with Feature Engineering (FE) techniques to identify and classify multiple types of botnet attacks targeting smart home IoT systems. The research focuses on evaluating DL models, specifically Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM), and Autoencoders (AE), using the BoT-IoT dataset. To manage the high dimensionality of the data, various features selection (FS) and reduction methods are applied, including Principal Component Analysis (PCA), Random Forest (RF) selection, and a hybrid PCA-RF method. These techniques help isolate the most relevant attributes, allowing the models to concentrate on key features that improve classification performance. A comparative analysis was performed to examine how each DL model responds to different FE strategies. The results demonstrated that both RNN-based models and AE showed strong performance in identifying botnet-related anomalies, reinforcing their suitability for smart home intrusion detection. The study also underscores the importance of employing advanced feature engineering methods to strengthen botnet detection capabilities within IoT environments.