Intrusion Classification and Detection System Using Machine Learning Models on NSL-KDD Dataset

Intrusion detection systems are crucial to cyberattack protection. This paper presents an intrusion detection system (IDS) architecture that uses numerous machine learning models to increase sensitivity and specificity. Popular models, including linear support vector machine, quadratic SVM, K-nearest-neighbor (KNN), linear discriminant analysis, multi-layer perceptron, LSTM, and auto encoder are tested on the NSL-KDD dataset. Our research measures these models’ accuracy, precision, recall, and F1-score to assess their ability to identify normal and malicious network data. We employ the algorithms’ capacity to recognize minor cyberattack patterns to construct a powerful defense. We investigate several methods to find the best machine learning models for intrusion detection. Our study shows the pros and cons of each model and how they assist in identifying intrusions. This research will assist in choosing IDS machine learning models to enhance network security and countermeasures. Experimental accuracy” measures the model’s prediction accuracy and is named after “accurate.” The K-nearest-neighbor algorithm is most accurate at 98.55%. The multi-layer perceptron method (97.78%) and long short-term memory algorithm (97.77%) follow closely.