Exploring the Performance of Machine Learning: A Case Analysis of Website Defacement Detection

Website defacement attacks are a prevalent form of cyber vandalism in which attackers alter a website’s visual appearance and content. These attacks typically involve unauthorized access to a web server, allowing attackers to replace legitimate content with their messages, often including political statements and propaganda. The primary goal is to disrupt the website’s functionality and damage its reputation. Several solutions for detection using machine learning have been designed and developed. This study explores the performance of nine machine learning algorithms, including Rocchio, Boosting, Bagging, Naïve Bayes (NB), K-Nearest Neighbor (KNN), Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), and Logistic Regression (LR), as well as five deep learning algorithms, including Deep Neural Networks (DNN), Reccurent Neural Networks (RNN), Long-Short Term Memory (LSTM), Gated Reccurent Unit (GRU), and Convolutional Neural Networks (CNN). In this study, data was collected from various sources, including websites affected by deface and normal websites, then evaluated the performance of each algorithm. The performance evaluation of each algorithm considers speed during training and the prediction accuracy illustrated by the learning curve graph for loss and accuracy, along with values for False Positives (FP) and False Negatives (FN). These findings reveal that the SVM model recorded a training time of 0.795 seconds and a prediction time of 0.005 seconds, achieving an accuracy rate of 99.1% and a minimum detection error with 15 for FP and 20 for FN. Meanwhile, the LR model took 0.071 seconds for training and 0.006 seconds for prediction, with an accuracy rate of 98.7%, generating detection errors with 17 for FP and 33 for FN.