Exploring Efficiency of Character-level Convolution Neuron Network and Long Short Term Memory on Malicious URL Detection

Machine learning techniques, especially deep learning neuron networks have been increasingly applied to solve the problems relating to information security and cybersecurity. Malicious URL (Uniform Resource Locator) detection is one of these. It is considered as a binary classification in machine learning, in which a URL or website address is classed as malign or benign. In this work, we implement the experiments on two different datasets to explore the efficiency of three proposed character-level deep neuron networks: (1) CNN (Convolution Neuron Network) based on VGG-16 architecture (Visual Geometry Group), (2) LSTM (Long Short Term Memory), and a fusion of CNN and LSTM for malicious URL detection. The experimental results are promising, especially for the fusion scheme of LSTM and CNN, with above 96% for precision and 98% for recall.