Attention-Based BI-LSTM Model to Detect Botnet Attacks over Internet of Things (IoT) Environments
Swapna Thota, D. Menaka · International Journal of Electronics and Communication Engineering · 2024
Detecting an IoT-botnet attack involves monitoring network traffic, identifying unusual behaviour, and implementing security measures to prevent and mitigate the impact of the attack. These days, hackers use botnets, a network of computational devices, to illegally access distributed resources and launch cyber-attacks against the “Internet of Things (IoT)”. A variety of “Machine Learning (ML) and Deep Learning (DL”) techniques have recently been developed to identify botnet assaults in IoT networks. The six main stages of the proposed paradigm are Botnet Attack Mitigation, Feature Extraction, Feature Selection, and Data Augmentation. First, data cleaning and data normalization (min-max normalization) are used to preprocess the raw data that has been gathered. The “Synthetic Minority Oversampling Technique (SMOTE)” method is then used to enrich the pre-processed data to address the class imbalance problem. Then, the supplemented data retrieved characteristics like Measure of Dispersion (Skewness, Variance, IQR), Central tendency (Generalized mean, Winsorized mean, Median, standard deviation, and variance), and Information Gain. The best features are selected using the extracted features. CUGOA stands for Clan Updated Grasshopper Optimization Algorithm, a hybrid optimization model. The Grasshopper Optimization Algorithm (GOA) and Elephant Herding Optimization (EHO) are combined to create the proposed CUGOA model. Next, the DCNN, Attention-based Bi-LSTM, and optimized RNN are all included in the new ensembled-deep-learning model, which detects Botnet Attacks. The chosen optimal features are used to fine-tune the DCNN and Attention-based Bi-LSTM. The improved RNN model receives the output of DCNN and Attention-based Bi-LSTM as input. The final detected outcome regarding the presence/ absence of a botnet attack is acquired from the optimized RNN model, whose bias function is fine-tuned using the new Hybrid optimization model. Once the attacker is found to be present in the network, it is mitigated using the new Botnet Traffic Filter (BTF). Thus, the network becomes highly reliable. The proposed model outperforms existing models regarding “accuracy, sensitivity, specificity, and precision”.