BULGARIAN ACADEMY OF SCIENCES

Abstract Researchers all over the world have provided significant and effective solutions to detect malicious URLs. Still due to the ever changing nature of cyberattacks, there are many open issues. In this paper, we have provided an effective hybrid methodology with new features to deal with this problem. To evaluate our approach, we have used state-of-the-arts supervised decision tree learning classifications models. We have performed our experiments on the balanced dataset. The experimental results show that, by inclusion of new features all the decision tree learning classifiers work well on our labeled dataset, achieving 98-99% detection accuracy with very low False Positive Rate (FPR) and False Negative Rate (FNR). Also we have achieved 99.29% detection accuracy with very low FPR and FNR using majority voting technique, which is better than the wellknown anti-virus and anti-malware solutions.