Malicious URL Detection using Sequence to Sequence Integration and Deep Learning Models

This paper presented a novel approach to malicious URL detection, utilizing various Deep Learning techniques including Simple RNN, LSTM, and GRU. The experiments were conducted on a malicious URL dataset sourced from Kaggle. Initially, traditional deep learning algorithms such as RNN, LSTM, and GRU were applied, achieving accuracies of 92%, 94.5%, and 93%, respectively. Subsequently, the study advanced to integrate sequence-to-sequence (Seq2Seq) models with deep learning methods to further enhance accuracy. The methodology involved thorough data preprocessing, model construction, and training on a labeled dataset of URLs. The Seq2Seq Model with RNN yielded an accuracy of 92.2%, while the Seq2Seq models with LSTM and GRU significantly improved accuracies to 97.5% and 94.3%, respectively. The evaluation of these Seq2Seq models demonstrated substantial improvements in accuracy, highlighting the efficacy of integrating Seq2Seq models with deep learning methods. These findings underscore the significant advancements made in enhancing the accuracy of malicious URL detection through the integration of Seq2Seq models with deep learning methods.