Fundamental-Indicator-Driven Stock Price Prediction for NSE-Listed Equities: A Hybrid GEO–CatBoost Framework with SHAP Explainability

Abstract

No prior study has applied fundamental financial indicators to NSE equity price prediction using machine learning, and Golden Eagle Optimization has never been applied to stock price forecasting in any market. This study addresses both gaps through the Golden Eagle CatBoost Prediction Framework (GECPF) which combines Golden Eagle Optimization (GEO) with CatBoost gradient boosting to predict NSE equity prices using four leak-free fundamental indicators: Sales General and Administrative to Revenue, Capital Expenditure Coverage Ratio, Return on Equity and Quick Ratio. The GEO fitness function jointly optimizes hyperparameters and feature relevance while enforcing inter-feature non-redundancy. Applied to five NSE- listed equities (2012-2022) with stringent chronological partitioning, GECPF achieved RMSE of 0.0400, R² of 0.9595, and directional accuracy of 79.17%, outperforming all eight benchmarked approaches. Wilcoxon test confirmed statistical significance above the closest competitor (p = 0.002, 10/10 run wins). SHAP study identified Return on Equity and Capital Expenditure Coverage Ratio as dominant predictors based on DuPont decomposition and free cash flow valuation theory. Findings demonstrate that leakage correction and feature discipline contribute more to predictive reliability than architectural complexity, scoped as a proof-of-concept on large-cap NSE equities.