The Phishing Detection Algorithm Research Based On Meta-learning

Along with the rapid development of computer network, cyber crime has becomeincreasingly rampant. As one of the important form of cybercrime, phishing brought greatthreat to people’s personal safety and property security. In this paper, on the basis ofcomprehensive study of domestic and international phishing detection methods,Combination of feature selection and Meta-learning, Departure from reducing the spatialdimensions of the data set, improve operational efficiency, improve the detection accuracyand generalization ability of the classifier, launched a study of phishing detection methods.Firstly, there are too many unrelated features and redundant features in existingphishing detection algorithm. Starting from to improve the base classifiers detectionaccuracy and increasing differences between classifier, this paper proposed ReMitraalgorithm which based on the fusion of the Relief and Mitra. For detection phishingshould feature selection preprocessing before classification. Relief algorithm calculatedthe differences between the feature of the same class and different class, delete unrelatedfeatures. Mitra algorithm calculated the correlation between features, remove redundantfeatures.Secondly, for the current phishing detection system accuracy is not high, Theproblem of false positives and false negative rate is too high. In order to improve theclassification accuracy of detecting, From the idea of ensemble learning, Proposed aphishing detection algorithm based on the meta-learning. The primary classifier selectionalgorithm based on support vector machine. From building large differencesclassifiers,processed the training set before the classifier training on it. Based onBagging,select different training subsets from original set, then apply ReMitra on thetraining set, select different feature sets. The combination of base classifier usingmeta-learning. Meta-level SVM learn the results of based classifier, producemeta-classifier and output final result.Finally, experiment to verify the proposed method in this paper,compared with thecurrent phishing detection, and looking forward to the future work.