Sentiment-driven deep learning framework for insider trading detection in Indian stock market

This study presents a novel approach for identifying insider trading in the Indian stock market by integrating sentiment analysis of financial news into deep learning models, specifically single-channel convolutional neural network (1CH-CNN) and multichannel convolutional neural network (MTC-CNN). Utilizing samples of insider trading and non-insider spanning from 2001 to 2020, along with corresponding financial news from the same period, we assess the effectiveness of the proposed approach across various time windows (30, 60, and 90 days) and its capability in predicting market manipulation. Our experimental results demonstrate that models incorporating sentiment metrics outperform those without, particularly in longer time windows, exhibiting enhancements in accuracy, precision, recall, F1-score, and ROC AUC. Notably, the integration of sentiment metrics results in a 20% reduction in the false positive rate across all time windows. These findings underscore the potential of sentiment analysis in augmenting insider trading detection mechanisms, emphasizing its importance for market surveillance and investor protection within the Indian stock market.