Analyzing of Classification and Feature Extraction Techniques: A Case as Phishing URLs

This paper presents a study on the classification of phishing URLs using machine learning. In this study, three classifiers (Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Deep Neural Network (DNN)) and three feature extraction techniques (Embedding layer, Long Short-Term Memory (LSTM), and Embedding layer with LSTM) were applied, using a dataset from Kaggle for training, validation, and testing. In the first phase of the study, the DNN with a combination of the embedding layer and LSTM showed the highest accuracy. This model was then modified by changing the configuration, resulting in five different configurations. The experimental results showed that Configuration 5 was the best option. This result was confirmed by verification with other datasets. Therefore, this model is the most appropriate in this study for screening or filtering phishing URLs or illegitimate links. In future work, it might be applied to implement an application for phishing link filtering.