Classifying Fraud in Indonesia Short Message Service (SMS) Using Machine Learning and Deep Learning

In the era of internet-based communication, Short Message Service (SMS) remains a critical platform for businesses and personal communication, particularly in Indonesia where smartphone penetration is high. However, the increase in SMS usage has led to a rise in fraudulent activities, posing significant risks to users. This research aims to enhance SMS classification in Indonesia by utilizing advanced machine learning and deep learning techniques. This research compares and investigates the effectiveness of various models, including Hybrid CNN-LSTM, Naive Bayes, Support Vector Machine (SVM), and advanced models like IndoBERT. Our approach involves classifying SMS into three categories: Fraud, Advertisement, and Normal messages, using a dataset of 1143 Indonesian SMS. The results demonstrate that combining several traditional machine learning methods into ensemble learning can significantly improve classification accuracy. Our hybrid model approach, especially the stacking ensemble learning combining CNN-LSTM, Support Vector Machine (SVM), and Naive Bayes, shows the best performance with over 94% score for accuracy, precision, recall, and F1-score. This research shows the potential of hybrid models in enhancing the accuracy and reliability of SMS classification, therefore contributing to a more secure and user-friendly SMS communication in Indonesia.