Empirical evaluations of machine learning and deep learning models for stock market predictions in Indian large cap equities: The limits of technical indicators and cost-aware back testing

Purpose: - The purpose of this study regarding the application of machine learning (ML) and artificial intelligence (AI) models to stock market prediction is in India with an interest in Large Cap Stock Equities and suitable evaluation strategies that are robust, reproducible, and cost aware in emerging markets. Design/methodology/approach - A good end-to-end research pipeline was developed with 15 years of daily data for 25 stocks listed in the NSE from yahoo finance, feature engineering using technical indicators, supervised classifiers with hyper parameter tuning, class balancing, calibrated thresholds and the use of advanced sequence models like Bidirectional LSTM. Evaluation included cost conscious back testing and intense statistical validation against benchmark strategies. Findings - we can conclude that out-of-sample accuracies and risk-adjusted return of standard technical indicators are not statistically different from naive buy-and-hold benchmarks, and combined with traditional models of ML and deep learning, these heuristics provide marginal predictive value. Feature importance analysis suggests that volume ratios, volatility, and short-term momentum signals are some of the predictors that are most relevant, though these are still not enough to always outperform market noise and market execution costs in the India large-cap space. Originality - The findings raise the importance of replicable, low-cost benchmarking and call for continued improvements in data, methodology and grounded real validation for ML & AI-driven financial forecast methods in emerging markets.