An English-Swahili Email Spam Detection Model for Improved Accuracy Using Convolutional Neural Networks

Abstract

E-mail has become an essential tool for digital communication, facilitating global networking and information exchange. However, spam emails, particularly those in multilingual contexts, pose a significant threat to cybersecurity. In 2023, cyber-related attacks cost Africa approximately USD 10 billion, with the Kenyan economy suffering losses of USD 383 million, 45% of which resulted from phishing and spam emails. While spam detection has been extensively studied for English, low-resource languages such as Swahili lack sufficient research and datasets. Swahili is spoken by about approximately 200 million people, mainly from East Africa. The same speakers use English as a medium of communication. This, therefore, highlights the need to research English-Swahili spam detection. This study recommends a convolutional neural network (CNN)-based model to increase spam detection accuracy in English-Swahili emails. The dataset comprises 8,829 ham emails and 2,749 spam emails, totalling 11,578 messages. The model was trained and evaluated via accuracy, precision, recall, and F1-score metrics. The results indicate a 99.4% accuracy rate, 99.3% precision, 98.2% precision, and 98.7% F1 score. These findings demonstrate good performance and effectiveness.