A Machine Learning Approach With PSO for Intrusion Detection

With the advent of internet and emergence of a digital landscape in the present world, sharing information online is prevalent. While this has made life much easier by facilitating speedy responses and widespread connectivity, it comes with its own set of challenges, the most concerning being threats to network security. The incessant threats to confidentiality of the information calls for measures to protect sensitive data. Network Intrusion Detection Systems have proved to be fruitful in monitoring malicious activities and thereby ensuring security of data by preventing unauthorized access in the systems. In this study, multiple machine learning classifiers were utilized to analyze the NSL-KDD dataset for network intrusions. DoS and Probe subsets were used and attempts were made to boost the performance of the machine learning models. Addressing class imbalance problem with the aid of SMOTE was crucial in stimulating unbiased model training. Machine learning algorithms like Support Vector Machines, K-Nearest Neighbors, Random Forest, XG Boost and Decision Trees were employed to train the models. The accuracies of the models were compared before and after implementing Particle Swarm Optimization for hyperparameter tuning.