Real-Time Fraud Detection in Telecommunication Systems Using Natural Language Processing and Advanced Machine Learning Algorithm

Scammers have siphoned away over \$1.03 trillion globally in the past year, emphasizing the urgent need for effective fraud detection systems. Fraud detection in telecommunication systems remains a significant challenge as fraudulent activities constantly evolve, resulting in financial losses and security risks. This paper proposes a fast and efficient machine-learning-based fraud detection system that analyzes phone call transcripts using Natural Language Processing (NLP) techniques. The models implemented include Naïve Bayes, Random Forest, and Support Vector Machines (SVM), utilizing BERT embeddings and TF-IDF vectorization for robust feature extraction. Data balancing using SMOTE and an ensemble-based classification approach further enhance the system's effectiveness.

Our proposed system achieves an impressive accuracy of 98.65\%, demonstrating its superiority over existing methods. Compared to other studies, this model offers enhanced accuracy, speed, and privacy-conscious design by eliminating the need for voice data making it a highly reliable and scalable solution.