Enhancing IOT Security: Hybrid Machine Learning for Detection of Botnet Attacks

Botnet attacks could well be regarded as one of the most perilous threats today, accompanied by a fast-evolving list of technical and operational challenges arising from issues related to network security. Botnet detection is therefore becoming more and more difficult; their exploitation techniques may change almost in real-time, and newer malware variants come out almost at breathtaking speed. Some good percent of new appliances implemented into the areas have been converted into an easy target for botnet infiltration. This scenario would imply a blanket of destruction and economic damages across a number of areas, put heavily forth by the resultant Interconnectivity-Internet of Things. Hence, the research has proposed a hybrid machine learning approach for a more effective detection of botnet attacks in the IoT setting. The new approach integrates the ANN-CNN-LSTM-RNN in a stacking algorithm called ACLR. The performance of the proposed model was validated through some benchmark tests against the individual performance of each of the models considered. Training and testing were performed on the UNSW-NB15 dataset containing nine attack types: Normal, Generic, Exploits, Fuzzers, DoS, Reconnaissance, Analysis, Backdoor, Shell Code, and Worms. The test metrics chosen to evaluate performance were: accuracy, precision, recall, and F1-score. Experimental evaluation shows that in most test scenarios, the performance of the ACLR model outperformed the individual models considered, thereby also achieving a higher accuracy, hence complementing the improved detection efficiency. This significantly boosts the possibility of enhancing botnet detection efforts in IoT systems to make an even stronger and more scalable countermeasure against cyber challenges.