Decoding Stock Market Predictions: Insights from Explainable AI Using Layer-wise Relevance Propagation

The growing use of AI models in financial markets demands both accuracy and transparency to build trust in highly regulated environments. Explainable AI (XAI) addresses the "black-box" nature of AI, providing insights into model decision-making, which is crucial for investors and regulators. This study builds on our previous work on market stability prediction using Layer-wise Relevance Propagation method to interpret our Conv1D-BiLSTM model, achieving 83% accuracy. Our experiments revealed that a 10-day look-back period optimally predicts one-day market stability. The outcome explanation with LRP highlights moving averages of local and foreign investors and open interest rates as the key economic indicators driving the AI model’s predictions. These insights deepen our understanding of market dynamics, particularly in the Pakistan Stock Exchange. We validate the robustness and scalability of our XAI framework, offering actionable insights for short-term investors and policymakers.