A Primary Dataset for Bangla Vishing Detection in Mobile Financial Services

The proposed dataset, "A Primary Dataset for Bangla Vishing Detection in Mobile Financial Services," comprises 3,262 transcripts of Bangla mobile phone conversations that have been manually examined. These were gathered to represent actual vishing attempts in services like bKash, Nagad, and Rocket. They include both authentic (2,000 samples) and fake (1,262 examples). Along with raw text, preprocessed text, tokenized representation, word IDs, and token counts, each entry is classified as either vishing or non-vishing. To ensure linguistic authenticity and sociocultural relevance, the dataset is characterized by brief, conversational snippets (average length ~8.5 tokens), which are frequently code-mixed and context-dependent. This dataset offers the first specific resource for Bangla vishing research, in contrast to earlier studies that depend on secondary or non-Bangla data. This allows for the creation of interpretable machine learning and natural language processing pipelines for fraud detection in low-resource environments.