International Journal of Science, Engineering and TechnologyAuthors: Research Scholar N. Rajarajeswari, Assistant Professor A. Subha
Abstract: Retail investors often display behaviourally induced biases such as overconfidence, loss aversion, and herding tendencies, which result in poor financial choices. Existing frameworks lack the capability of modelling the complex and contextual nature of these biases. In this paper, we suggest a framework for modelling and interpreting behaviourally induced biases using XML through trading logs of 5,000 retail investors. In this work, we apply Random Forest (RF), XGBoost, and a customized hybrid attention-based neural network, along with SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) methods to achieve interpretability. Our approach obtains an accuracy rate of 89.2% in bias prediction compared to logistic regression (74.5%). Our results show that the highest contribution to bias comes from loss aversion and recency. Comparison studies prove the superiority of our proposed framework over existing frameworks in terms of transparency.
