High Efficiency Spam Filtering: A Manifold Learning-Based Approach

Abstract

Spam filtering, which refers to detecting unsolicited, unwanted, and virus-infested emails, is a significant problem because spam emails lead to unnecessary costs of Internet resources, waste of people's time, and even loss of property. Support vector machine (SVM) is the state-of-the-art method for high accuracy spam filtering. However, SVM incurs high time complexity because of the high dimensionality of the emails. In this study, we propose a manifold learning-based approach for time-efficient spam filtering. From the experiments that most of the features are not decisive, we can obtain the viewpoint that only a minor part of the spam emails can be detected using the nondecisive features. Based on the insight, we propose to employ the Laplace feature map algorithm to obtain the geometrical information from the email text datasets and extract the decisive features. Then, the extracted features are used as the input of SVM to spam filtering. We conduct extensive experiments on three datasets, and the evaluation results indicate the high accuracy time efficiency of our proposed algorithm.