Binary and Multiclass Classification Intrusion Detection System using Benchmark NSL-KDD and Machine Learning Models

In the modern world, cyber security is growing in importance. Every day, there are more and more people using the internet, and they are downloading vast amounts of data from various websites onto their gadgets. Through the introduction of various attack types, attackers aim to obtain entry to the systems of regular users. To defend regular users against intruders, a multitude of intrusion detection systems are being created. Since most of these systems were created using obsolete datasets, there is room for improvement in terms of detection rate, accuracy, and false alarm rate. The suggested system builds various shallow machine learning frameworks for binary as well as multiclass categorization. Proposed system applies shallow learning techniques like Random Forest, Decision Tree, Gaussian Naïve Bayes, Support Vector Machine, K-Nearest Neighbor, Gradient Boost, AdaBoost and Linear Discriminant Analysis on benchmark NSL-KDD dataset. Suggested system selects best features by applying SelectKBest and Correlation Features Selection methods on NSL-KDD dataset. All machine learning frameworks are build using selected features KDDTrain+ dataset. Grid-search is applied to get the optimal hyperparameters of a models which helps in the most accurate results. Testing is carried out on distinct test dataset KDDTest+. It is observed that the performance of Gradient Boosting model for binary classification is highest as compared with other models where, Accuracy- 86%, Precision-89%, Recall- 84%, and F1 score- 85% Also, performance of Gradient Boost and Random Forest model is better for multiclass classification, in comparison with other shallow learning models where, Accuracy- 90%, Precision- 92%, Recall-90%, and F1 score- 90%.