Content Based Approach for Detecting Smishing Messages in Mobile Phones Using an Improved Convolutional Neural Networks Model

Abstract

SMS stands for Short Message Service (SMS). Short messaging service is a text messaging service where a user can send short messages via a mobile device. Short message service has evolved and become very popular as a communication medium in the last decade. It has become a more effective mode of communication compared to email. Unfortunately, smishing (SMS phishing) has emerged as the most common type of spam because traditional detection methods have difficulty understanding the informal nature of these messages. An improved class of CNN-based models targeted at accurate detection of smishing on mobile devices was developed. Deep learning theory was used in this work. (UCI) refers to University of California, Irvine (UCI). The UCI Machine Learning Repository contains datasets, domain theories, and data generators used by the machine learning community to empirically study machine learning algorithms. In this study, a research design was carried out and samples of the UCI Machine Learning Repository were used to build an experimental model. The analyzed dataset was a set of 5, 574 SMS messages from both spam and non-spam messages. The performance metrics used were Precision, Recall, F1-Score and Accuracy. The CNN model used for evaluation, had a bigger number of hidden layers for better detection. A higher accuracy of 99. 95% was achieved, indicating good performance and better detection of the SMS spams (SMS phishing). In the analysis mentioned in the text, the text preprocessing greatly contributes to improved detection accuracy and CNN outperforms the traditional detection methods. It shows that the sophisticated nature of smishing attacks make it necessary for advanced detection mechanisms to be applied to prevent future SMS threats. Although some authorities for implementation should allocate resources to implement these solutions, other authorities must define roles for different detection systems in order to realize the ideal and continuous performance of the detection tools. However, various authorities should also continuously enhance detection mechanisms by feature enhancement, data augmentation, and regular performance evaluation. The training for the staff and establishment of the performance benchmarks needs to be implemented. The users should also contribute with their comments, reports suspicious message, and raising awareness for each other, while all other stakeholders should make available their expertise and resources to support this work.