Lightweight intrusion detection system for IoT with improved feature engineering and advanced dynamic quantization
Semachew Fasika Misrak, Henock Mulugeta Melaku
Abstract
In recent years, IoT devices have surged globally, enhancing businesses, industries, and daily life. Nevertheless, IoT devices are not immune to cyber attacks. On the contrary, their limited memory and processing power make them more susceptible to malicious attacks. Therefore, developing a lightweight yet efficient attack detection is a critical issue for IoT systems. This research proposes an efficient lightweight hybrid deep learning model (DNN-BiLSTM) to detect and classify attacks in an IoT system utilizing improved feature engineering and advanced quantization. Although leveraging hybrid deep learning model that combines DNN and BiLSTM facilitates the extraction of intricate network features in a nonlinear and bidirectional manner, aiding in the identification of complex attack patterns and behaviors, tailoring this for IoT devices remains challenging. To address the constraints inherent to IoT devices, this research incorporates improved feature engineering techniques, specifically Redundancy-Adjusted Logistic Mutual Information Feature Selection (RAL-MIFS) with a two-stage Incremental Principal Component Analysis (IPCA) algorithm. Additionally, advanced quantization techniques, including Quantization Aware Training (QAT) and Post-Training Dynamic Quantization (PTDQ), alongside advanced Optuna for hyperparameter optimization, are utilized to enhance computational efficiency without compromising detection accuracy. Experimental evaluations were conducted on the CIC-IDS2017 and CIC-IoT2023 datasets to assess the performance of a quantized DNN-BiLSTMQ model. The model demonstrated competitive detection accuracy and computational efficiency compared to state-of-the-art methods, including autoencoder + ensamble learning, LNN and CNN-BiLSTM. Using the CIC-IDS2017 dataset, a detection accuracy of 99.73% is achieved with a model size of just 25.6 KB, while on the CIC-IoT2023 dataset, the achieved a detection accuracy is 93.95% with a model size of 31.3 KB. These results highlight the potential of quantized DNN-BiLSTMQ model for efficient and accurate cyber attack detection on IoT systems.