Intrusion Detection Schemes Based on Synthetic Minority Oversampling Technique and Machine Learning Models
Ali Hussein Ali, Maha Charfeddine, Boudour Ammar, Bassem Ben Hamed · 2024
With the progression and sophistication of technology, the frequency and intricacy of cyber-attacks also escalate. Malicious hackers and cybercriminals are perpetually devising novel techniques to infiltrate computer systems and steal vital data illicitly. To tackle these risks, organizations must deploy efficient Intrusion Detection Systems (IDSs) to detect and respond to attacks promptly. In recent times, IDSs have garnered considerable interest due to their efficacy and accuracy in identifying anomalous patterns in network traffic through the utilization of machine learning methodologies. This study aims to develop a model capable of detecting intrusions by utilizing several machine learning algorithms on the selected features derived from the modeling procedure. Multicollinearity is a statistical method employed to identify high relationships between variables. This study uses a Synthetic Minority Oversampling Technique (SMOTE) to tackle the problem of imbalanced datasets. SMOTE is a powerful technique that can successfully balance the distribution of classes and improve the accuracy of a machine learning model in identifying the underrepresented class. A study is done to examine several machine learning approaches and evaluate the efficiency of the proposed methodology. The performance of Machine Learning algorithms on the well-established Intrusion Detection benchmark CSE-CIC-IDS2018 and KDD CUP99 datasets shows great potential.