Leveraging hybrid ensemble models in stock market prediction: A data-driven approach

Forecasting the stock market with precision remains a task because of the intricate and ever-changing nature of market data dynamics. This research seeks to enhance precision by harnessing ensemble machine-learning models that capitalize on the unique advantages of each model while addressing their drawbacks. This study addresses this gap by testing models across six major global markets (S&P 500, NASDAQ, DAX, FTSE, Nikkei 225, and Hang Seng), providing insights into the ensemble model's performance in diverse economic contexts. We employed gradient boosting, the decision trees method, the neural networks approach, and Bayesian ridge models to foresee stock prices. We merged them through voting and stacking strategies. The models were assessed using mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE) and the a20 index. Analysis involving analysis of variance (ANOVA) and subsequent Tukey's honest significant difference (HSD) tests was performed to validate performance variations among the tested models. In the global market, the stacking ensemble showed the lowest MAE of 0.0332, followed closely by the voting ensemble, which achieved 0.0347 MAE. These results hint that ensemble models are more accurate in making predictions than standalone and hybrid models in stock price prediction tasks.