Detection of malicious URLs with deep learning

In recent years, with the increasing use of the internet, it has become a crucial aspect of our lives. New communication technologies, social networks, e-commerce, and various applications, including online banking and the use of technological home appliances, play a significant role in promoting and expanding business. The purpose of this thesis is to detect malicious URL addresses using the artificial intelligence model employed, working with a large dataset to achieve the best results. Existing studies in the literature were reviewed in this thesis, emphasizing the contribution of dataset size to accuracy success rates. In the study, the high number of URLs in the data set and the use of RNN model architecture increased the performance values by 2 points. A 7-layered RNN model was created in this work, combining two similar national and international datasets to form a massive new dataset comprising 579,112 URL addresses. This new dataset was then divided into training and test sets. Initially, the model was trained on the dataset and subsequently tested on the second dataset. When this dataset was processed through our model, 91% was achieved, indicating results in detecting malicious URL addresses. This study aims to make a significant contribution to the development of more effective methods for detecting malicious URL addresses.