Malsmsdetector: malicious text message detector with hybrid feature vector and stacked ensemble model: a comparative study

In recent years, the emergence of telecommunication systems has led to an increase in global electronic messaging traffic. Most of this traffic contains unwanted content for the user. In this study, an approach is proposed in which feature vectors generated using DBOW and PV-DM techniques are used for classification as a hybrid for spam SMS detection. In the training and testing of the proposed method, four different datasets (UCI, BEC, Big NUS and DITNUS) that are widely used are combined and used. This dataset is tested with 10 different machine learning algorithms and then a unique stacked ensemble model is proposed to increase the performance. In the tests using the model, accuracy, precision, recall, F-score and AUC values are 98.38%, 98.39%, 98.39%, 98.37% and 96.81%, respectively. When 10-fold cross validation is applied to the obtained results, the standard deviation value is 0.004. The analysis time per sample is 0.087 milliseconds.