Stock Market Trend Prediction using an Explainable Deep Learning Approach

Market trend research is essential for taking advantage of the best investing opportunities to maximize profits and the quickest way to minimize losses, because stock forecasting is complex and fraught with dangers and uncertainties. All five of the following stock-market trends—upward, downward, double-top, and rounded-top—can be predicted using our proposed deep learning model. Average accuracy for the suggested model is 94. 9 \%, which is higher than common benchmarks like logistic-regression (53. 41%), random-forest (85. 14 \%), and support-vector-machine (62. 56). In addition, across four real-world heterogeneous datasets, the suggested model outperforms the competition in terms of F1-score, with a performance of 94. 85 \%. In comparison, RF achieved 77. 95 \%, SVM 21. 02 \%, and LR 46. 23 \%. In order to make the forecasts more understandable to stakeholders, we also use SHAP and LIME, two XAI approaches, to improve interpretability. By identifying the ten most significant characteristics, the SHAP analysis allows for feature reduction without sacrificing performance. This study's findings show promise for real-world use in financial decision-making by helping investors with risk management and long-term planning through a combination of interpretability and predictive power.