Hybrid Machine Learning Model for Efficient Bonet Attack Detection

With the rapid development of Internettechnology, cyber-attacks are becomingincreasingly sophisticated, with botnet attacksemerging as one of the most harmful threats.Botnet identification is challenging due to thewide range of attack vectors and the continuousevolution of malicious software. As the Internetof Things (IoT) technology expands, manynetwork devices are susceptible to botnet attacks,leading to significant losses in various sectors.This paper proposes a botnet identificationsystem using a Long Short-Term Memory(LSTM) model, a popular deep learningapproach, to effectively distinguish betweennormal network traffic and botnet attacks. Themodel classifies network traffic into twocategories: normal (0) and botnet attack (1).Experiments were conducted using the UNSWNB15dataset, which contains nine types ofattacks, including ‘Normal’, ‘Generic’,‘Exploits’, ‘Fuzzers’, ‘DoS’, ‘Reconnaissance’,‘Analysis’, ‘Backdoor’, ‘Shell code’, and‘Worms’. The LSTM-based model achieved animpressive testing accuracy of 90%. Theproposed approach demonstrates strongperformance in identifying botnet activities, withhigh receiver operating characteristic (ROC) areaunder the curve (AUC) and precision-recall areaunder the curve (PR-AUC) scores, indicating itseffectiveness in classifying normal and attacktraffic. Performance comparisons with existingstate-of-the-art models further validate therobustness of the proposed LSTM-basedapproach. This research contributes to enhancingcybersecurity procedures by providing a reliabletool for detecting botnet attacks in evolvingnetwork environments.