Hybridized Deep learning Techniques for Enhanced SMS Spam Detection system

Abstract

Short Message Service (SMS), popularly call text messaging, has revolutionized communication by enabling rapid and convenient information exchange among users.  Despite its widespread use it comes with some flaws that has made it a target for spanners. This justifies the need for spam detection system. The study developed an hybridized spam detection system. The dataset used for the spam detection classification was downloaded from kaggle .com repository. Two Feature extraction (FE) which are: Term Frequency-Inverse Document Frequency (TF-IDF) and Bidirectional Encoder Representations from Transformers (BERT) were used. The study then employed three techniques which are LSTM, CNN -LSTM and Linear Regression. The results of the three-model developed for spam detection revealed that CNN-LSTM model achieves the highest ACC (99%), followed by LSTM (98%) and Logistic Regression (94%). CNN-LSTM also recorded superior performance in precision (98%), Sen (91%), and F1-score (94%). The study concluded CNN-LSTM achieves state-of-the-art accuracy, while LSTM also demonstrates strong performance. Logistic Regression, while providing a good baseline, is generally outperformed by the deep learning approaches. The model is recommended for mobile communication sector to protect privacy violation of the user. More deep learning techniques and FE can be employed in future in order to increase the ACC of the model.