The curse of 140 characters: evaluating the efficacy of SMS spam detection on android

Abstract

Many applications are available on Android market place for SMS spam filtering. In this paper, we conduct a detailed study of the methods used in spam filtering in these applications by reverse engineering them. Our study has three parts. First, we perform empirical tests to valuate accuracy and precision of these apps. Second, we test if we can use email spam classifiers on short text messages effectively. Empirical test results show that these email spam classifiers do not yield optimal accuracy (like they do on emails) when used with SMS data. Finally, in this work we develop a two-level stacked classifier for short text messages and demonstrate the improvement in accuracy over traditional Bayesian email spam filters. Our experimental results show that spam filtering precision and accuracy of nearly 98% (which is comparable with those of email classifiers) can be obtained using the stacked classifier we develop.