A Novel Phishing Detection Research Based On Revised Multi-Objective Evolution Optimization Algorithm And Random Forest

Phishing attacks have become a major threat to people’s daily network environment.Phishing attackers disguise trusted websites to defraud users’ trust and steal sensitive data of users,causing losses to users.Therefore,an effective method is needed to prevent phishing attacks from causing sustained losses to people.The machine learning models are widely used in many phishing detection systems for classifying the massive phishing datasets.Based on the experiences,researchers prefer to extract as many features as possible to improve phishing detection performance.However,redundant and useless features in the feature set will degrade the performance of the underlying classification models.Meanwhile,many of the existing phishing detection models are mainly focus on the detection accuracy with the recall rate underappreciated.However,in the phishing detection,it is more harmful to detect a phishing website as a legitimate website than to detect a legitimate website as a phishing website.This dissertation proposes the MOE/RF,a novel phishing detection model based on the revised Multi-Objective Evolution optimization algorithm(MOE)and Random Forest(RF).The MOE/RF model not only takes the accuracy as the detection target,but also minimizes the probability of false detection of phishing sites as legitimate ones.Meanwhile,two new strategies,the symmetric uncertainty-based population initialization and the population state-based adaptive environmental selection,are proposed to improve the performance of the MOE.Experimental results on testing three different phishing datasets have demonstrated that the MOE/RF performs better than many of the existing methods.The main contributions of this dissertation are described as follows:1)Apply multi-objective evolutionary optimization algorithm to phishing detection.Traditional phishing detection methods always ignore the harmfulness of mistakenly detecting phishing websites as legitimate websites,and often only focus on improving the accuracy of the model.In MOE/RF model,we not only consider the accuracy of the model,but also minimize the probability of false detection of illegal websites as legitimate websites to improve the performance of the model.2)Design a population initialization strategy based on symmetric uncertainty.In MOE/RF model,symmetric uncertainty is used as an index to measure the importance of fishing features,which can evaluate the correlation between a single feature and website tags.Compared with the random population initialization strategy,using the prior knowledge of SU in the population initialization process can effectively avoid the invalid search of MOE.3)Design an adaptive environment selection strategy based on population state.According to the distribution of non-dominant individuals,the strategy divides the population into three states in the evolutionary process.Based on the newly defined three states,an adaptive environment selection strategy is proposed,that is,different selection strategies are adopted for different population states.Experiments show that this strategy ensures the convergence and the diversity of the model.4)Random forest is used as the basic classifier and the model MOE/RF designed in this dissertation is compared with the latest fishing detection model and traditional fishing detection model.Experiments verify the rationality of using random forest as classifier,and verify the effectiveness of the model designed in this dissertation.