A Hybrid Gradient Boost Model for Intrusion Detection

Abstract

Due to the advancement of network threats at present, it is crucial to conduct research on identifying and preventing network anomalies. Machine learning (ML) is one strategy for Intrusion Detection System (IDS). Finding a reliable system to act as a networking shield is still difficult despite the fact that various IDS are suggested utilizing ML. This paper suggests a hybrid model combined with the gradient boost methods XGBoost and Lightgbm to forecast the various attacks that are urging in the network. To obtain the higher precision, hyperparameters of the algorithms are tuned. The proposed system is trained using the UNSW-NB15 dataset, which contains attacks for Generic, Exploits, Denial of Service (DoS), Shellcode, Fuzzer, and Reconnaissance. The system has an average accuracy of 99.89%. Because of the recent dataset training, the proposed system is relevant to modern Intrusion Detection Systems used in current network systems.