Comparative Analysis of Deep Learning Models for Web Defacement Detection Based on Textual Context

Web defacement refers to the unauthorized alteration of the appearance or content of a website, which can have detrimental effects on its operations and financial standing. Deep learning techniques have shown promise in detecting web defacement incidents. This research explores the application of three deep learning models, namely Bi-LSTM, GRU, and BERT, for web defacement detection. The study collects data from various sources, including defaced and regular websites, followed by preprocessing steps such as case folding. The selected models are then employed to build the detection models. Evaluation metrics such as accuracy, precision, and F1 score are used to compare the performance of the models. The findings reveal that the Bi-LSTM model outperforms the GRU and BERT models in terms of accuracy (0.990), precision (0.984), and F1 score (0.953) for classifying web defacement based on textual content. This superiority can be attributed to the Bi-LSTM model’s ability to process sequential information and capture contextual dependencies using the recurrent neural network architecture. It leverages bidirectional information flow, enabling better understanding and representation of the text. However, it is essential to consider other factors impacting model performance, such as dataset characteristics, data preprocessing techniques, and model configurations. The results of this research provide valuable insights into utilizing deep learning models, particularly Bi-LSTM, for effective web defacement detection.