Domain-Adaptive Transfer Learning for Privacy-Preserving Scam Call Detection

Created using ChatSlide

This research explores innovative strategies to combat vishing — a form of voice phishing — by addressing privacy concerns, the lack of annotated datasets, and latency issues in detection. Utilising a composite dataset, a two-stage training approach is developed and tested through offline implementation, achieving notable accuracy and revealing insights via SHAP analysis. Despite challenges such as ASR accuracy dependency and generalisation issues, the study proposes future uses of federated...