Phishing Detection: Malicious and Benign Websites Classification Using Machine Learning Techniques

Malicious Web Links Detection - A Comparative Analysis of Machine Learning Algorithms

“…Comparing the last step of the experiments, about feature importance, we managed to confirm the results within other articles. The experiments done in [11] concludes that no particular features dominated the detection algorithm. On the contrary, Johnson et al [8] manage to prove the relevance of URL related attributes through chi-squared test, as our lexical URL features achieved a medium accuracy score.…”

“…Johnson et al [8] is experimenting with multiple algorithms, including KNN, which scores 97.47 accuracy for the binary classification. In [11] and [10], KNN proves to be the most flexible and adaptive model across multiple datasets. Pakhare et al [12] uses KNN in the best performing ensemble model, including SVM and DT.…”

“…Some approaches consider a binary classification (most approaches) and others a multi-class problem ( [8,16]). There are articles ( [10], [11]) testing the proposed models across multiple datasets to prove their adaptability.…”

“…The classification done by Oshingbesan et al [11] is binary and it is an experiment across multiple different datasets with multiple artificial intelligence algorithms (SVM, DT, RF, KNN etc.). According to [11], KNN is the most robust model that achieved a high performance in a cross-dataset environment. Similarly, Naveen et al [10] is implying usage of tree-based models (DT, RF) and other algorithms (LR, Linear SVM, KNN etc.)…”

“…Johnson et al [8] experiments with DT for binary classification and achieves an accuracy of 97.63. For Oshingbesan et al [11] and Naveen et al [10], DT performs well for classifying links based on lexical features but is not the best performing model. Tung et al [16] achieves a multi-class accuracy of 96% for the DT model.…”