Genetic Optimization Techniques for Enhancing Web Attacks Classification in Machine Learning
Ameera Jaradat, Ahmad Nasayreh, Qais Al-Na’amneh, Hasan Gharaibeh, Rabia Emhamed Al Mamlook
Abstract
Web-based applications are now the preferred approach for delivering a variety of services via the Internet. As a result of the globalization of commerce, web applications have been growing quickly and becoming increasingly complicated. Such applications have a significant security vulnerability in the online environment since they were developed with little experience and without testing or validation. Numerous attackers use this security vulnerability to take control of the program, modify the data, and steal the most crucial information. They may also access all internal, unauthorized items. This work presents a hybrid model that classifies website attacks as benign through the integration of four machine learning algorithms: gradient boost (GB), multilayer perceptron (MLP), logistic regression (LR) and K nearest neighbor (KNN). The work deployed Tree-based Pipeline Optimization Tool (TPOT) that utilizes the Genetic algorithm (GA) to extract the optimal parameter and consequently enhance the model performance. The model underwent evaluation utilizing a data set from the Canadian Institute 2023 that contains various types of attacks. Among these algorithms, GB achieved the best accuracy scores of 95%, 94% and 95% for accuracy, recall and F1-score, respectively.