Advancing stock price prediction through the development of hybrid ensembles: a comprehensive comparative analysis of machine learning approaches

This study introduces an innovative ensemble learning approach for stock direction classification and prediction. The methodology enhances accuracy through strategic base learner selection, hyperparameter optimisation using GridSearchCV, and hybrid ensemble techniques. Leveraging a diverse set of machine learning models—including Logistic Regression (LR), Random Forest (RF), Decision Trees (DT), Support Vector Machines (SVM), K-Nearest Neighbors (KNN), AdaBoost, and Lasso-LSTM—this research compares single classifiers, hybrid ensembles, and a Voting-based classification model. Using S P 500 index data spanning from June 8, 2016, to September 6, 2023, the study evaluates classification performance across multiple metrics: accuracy, precision, recall, F1 score, AUC, and sensitivity. Statistical analysis, including Kruskal–Wallis tests and Dunn’s post hoc comparisons, reveals significant performance variations among models. The results indicate that the SVC + LR + RF + VOTING ensemble achieves the highest accuracy (95.8%) and AUC (0.97), followed closely by AdaBoost (ADAB) with an accuracy of 94.2% and AUC of 0.96. These findings underscore the superiority of ensemble classification methods over individual classifiers in stock market prediction. This research provides practical insights for financial decision-makers by addressing key gaps in the existing literature and emphasising the importance of diverse evaluation metrics in model assessment. The study establishes a solid foundation for classification-based predictive modelling in stock market analysis and encourages further research into diverse datasets and advanced classification algorithms to enhance model generalisation.