Enhancing Intrusion Detection Systems through Advanced Machine Learning Techniques

Abstract

The rise in cyberattacks during the Covid – 19 pandemics has intensified the need for intrusion detection systems. This study investigates three machine learning algorithms: Random Forest Classifier, Convolutional Neural Network – Focal Loss, and Convolutional Neural Network – Cross Entropy Loss, under two project development lifecycles. Using the NSL – KDD Dataset, which contains network connection records classified into normal traffic or a multitude of different attacks, each of system is evaluated across multiple data handling staged, including scaling, balancing and feature selection. Results indicate that the Random Forest Classifier is best for binary classification, no matter the project development approach used (F1 Score: 99.92% in both cases); whereas Convolutional Neural Network – Cross Entropy Loss works best for multi – class classification when under the micro – detectors approach (F1 Score: 99.59%). As such, the findings of this research would offer practical guidance for designing machine learning systems for intrusion detection systems. Key Terms – Convolutional Neural Networks, Disposable Micro – Detectors, Intrusion Detection Systems, and Random Forest Classifier.