Evaluating the Performance of Long Short-Term Memory for Web Attack Detection

Web threats provide a continuing and dynamic hazard to both individuals and organizations, encompassing a wide spectrum of malicious activities such as SQL injections and denial-of-service attacks. In this research, we look into how Long Short-Term Memory (LSTM) neural networks can be used to better detect malicious activity on the web. LSTM is a subset of Machine learning algorithms and it’s a potential tool for web attack detection because of its abilities in sequential data analysis. Experiments were conducted utilizing the CIC-DDoS2019 dataset, which comprises a diverse range of Distributed Denial of Service (DDoS) attacks and benign network traffic. The multi-layered LSTM model we have developed demonstrates effective discrimination between normal and attack occurrences in the context of binary classification and multiclass classification. In order to optimize performance, we conducted experiments where we manipulated several model parameters, such as the number of layers, activation functions, sample size, and epochs. The evaluation metrics included in our study are accuracy, precision, recall, and F1-score. We obtained an impressive accuracy rate of 99.7 % for portmap attacks, 99.9% for LDAP attacks, and 99.9% for NTP attacks. For multiclass classification, we obtained 99.9% accuracy rate. The result can be attributed to the structure of the model, the utilization of high-quality training data, and the optimization of hyperparameters.