SMS Spam Detection Based on Long Short-Term Memory and Gated Recurrent Unit

An SMS spam is the message that hackers develop and send to people via mobile devices targeting to get their important information. For people who are ignorant, if they follow the instruction in the message and fill their important information, such as internet banking account in a faked website or application, the hacker may get the information. This may lead to loss their wealth. The efficient spam detection is an important tool in order to help people to classify whether it is a spam SMS or not. In this research, we propose a novel SMS spam detection based on the case study of the SMS spams in English language using Natural Language Process and Deep Learning techniques. To prepare the data for our model development process, we use word tokenization, padding data, truncating data and word embedding to make more dimension in data. Then, this data is used to develop the model based on Long Short-Term Memory and Gated Recurrent Unit algorithms. The performance of the proposed models is compared to the models based on machine learning algorithms including Support Vector Machine and Nave Bayes. The experimental results show that the model built from the Long Short-Term Memory technique provides the best overall accuracy as high as 98.18%. On accurately screening spam messages, this model shows the ability that it can detect spam messages with the 90.96% accuracy rate, while the error percentage that it misclassifies a normal message as a spam message is only 0.74%. Index Terms-SMS spam, natural language process, deep learning, long short-term memory