Automated SMS Classification and Spam Analysis using Topic Modeling

A huge amount of illegitimate short message service (SMS) messages flood users' mobile phone inbox and worsen users' experiences. This paper presents an automated framework for filtering SMS spam with the use of various classifiers. The text of SMS is preprocessed to generate a large number of tokens. These tokens individually or combinedly act as a feature and generate a large number of features. The comparative analysis of different machine learning algorithms with and without feature selection methods is performed. The classifiers are categorized into three groups based on their working namely ensemble-based, decision tree-based and others. The used classifiers are assessed using metrics such as accuracy, precision, recall, f-score and area under the receiver operating characteristics curve (AUC-ROC)). The random forest along with the chi-square feature selection method achieved the best performance. The major categories of SMS spam using topic modeling approaches such as latent Dirichlet allocation (LDA) and non-matrix factorization (NMF) are discussed.