Neurocomputational Modeling of Cognitive Biases in Sports Betting: A Machine Learning Approach

In recent years, studies on sports betting have emphasized the use of statistical models and rational choice frameworks. They prioritize prediction accuracy and profitability while often neglecting cognitive biases. Few studies have integrated specific biases into the computational models. Furthermore, existing machine learning approaches mainly detect outcomes rather than model decision processes. This study proposes a neurocomputational framework. It combines cognitive bias theory and deep learning to predict bias-driven betting actions. We designed an LSTM with attention to the capture temporal dependencies. These dependencies are shaped by the hot-hand fallacy, loss chasing, and confirmation biases. The experimental results showed that the standard evaluation yielded an F1-score of 0.462. However, optimizing the decision threshold significantly improved the performance to an F1-score of 0.636 (AUC = 0.692). This highlights the sensitivity of the bias detection. The proposed model demonstrated superior ranking capabilities compared to the optimized baselines. Moreover, the attention weights provide interpretable insights into the distinct memory horizons of different biases. This framework offers a novel tool for forecasting biased-betting behavior. It has applications in responsible gambling analytics and risk predictions.