A HYBRID MACHINE LEARNING MODEL FOR NETWORK INTRUSION DETECTION

O. O. Solanke; T. A. Adegboyega; A. I. Taiwo; K-K A. Abdullah; F. E. Ayo; T. J. Odule; S. O. Hassan; B. T. Efuwape

O. O. Solanke Department of Computer science, Olabisi Onabanjo University, Ago-Iwoye, Ogun State
T. A. Adegboyega Department of Computer science, Tai Solarin University of Education, Ijagun, Ogun State
A. I. Taiwo Department of Computer science, Olabisi Onabanjo University, Ago-Iwoye, Ogun State
K-K A. Abdullah Department of Computer science, Olabisi Onabanjo University, Ago-Iwoye, Ogun State
F. E. Ayo Department of Computer science, Olabisi Onabanjo University, Ago-Iwoye, Ogun State
T. J. Odule Department of Computer science, Olabisi Onabanjo University, Ago-Iwoye, Ogun State
S. O. Hassan Department of Computer science, Olabisi Onabanjo University, Ago-Iwoye, Ogun State
B. T. Efuwape Department of Computer science, Olabisi Onabanjo University, Ago-Iwoye, Ogun State

Keywords: Network Intrusion, Machine Learning (ML), AdaBoost, SGDC, ADA_SGDC, SGDC_ADA

Abstract

Intrusion detection is a significant challenge in network security, as it involves detecting unseen attacks in a network or system. In this research, we aimed to build a hybrid machine learning model for intrusion detection using artificial intelligence (AI). To do this, we used the KDD CUP 99 dataset and applied two machine learning algorithms: AdaBoost and Stochastic Gradient Descent Classifier (SGDC). These algorithms were combined to form two hybrid models: SGDC_ADA and ADA_SGDC. The results of our study showed that the SGDC_ADA model had an accuracy of 0.97 and outperformed the ADA_SGDC model, which had an accuracy of 0.96. In addition, the SGDC_ADA model had an average precision of 0.97, average recall of 0.96, and average F1-score of 0.97, while the ADA_SGDC model had an average precision of 0.96, average recall of 0.95, and average F1-score of 0.96. Overall, our research suggests that the SGDC_ADA hybrid model is an effective method for intrusion detection, with high accuracy and low error rates. This model may be useful in improving network security and protecting against unseen attacks.