A Novel Approach of Detecting Malicious Phishing URL Using Self-organizing Map

Empirical study attempts to analyze major trends of phishing Uniform Resource Locator detection, emphasized on improvement of 80% training data on feature extraction using term frequency-inverse document frequency (TFIDF) based on self-organizing map with classification machine learning algorithm. To analyze characteristics of phishing Uniform Resource Locator term frequency-inverse document frequency (TFIDF) is implemented for phishing feature extraction incorporated with self-organizing mapping (SOM). Significance of a phishing features in URL are subjected to data frame depending on quantity and quality of malicious attacks. For advanced classification techniques support vector classifier, Multinomial naïve Bayes, Random Forest Classifier, Extreme Gradient Boosting (XGB) Classifiers are compared using python programming language to determine best performance with proposed model. Essential comparison performance data used for risk prediction in order to select best classification model incorporated with proposed model. Proposed framework is able to achieve a superior accuracy with Kappa performance indicators of score 0.686 which considered as moderate value. Suggested method is based on URL features containing entropy, address bar-based features, IP-based features, and URL-based features. Investigation provides significance mean accuracy of 97.5% using Support Vector Classifier (SVC). Proposed model benchmark 10 datasets each containing 11430 URL (50% legitimate and 50% phishing) based on 87 features. Other classifiers produced an accuracy rate of Multinomial naïve Bayes (84% accuracy), Random Forest Classifier (93% accuracy), Extreme Gradient Boosting (XGB) Classifier (96% accuracy). Study limits the use of more phishing features in the dataset.