Stock Market Forecasting Using Hyperparameter-Tuned Ensemble Model

Stock market forecasting is fascinating due to the financial interests associated with it. The conventional method uses Technical Indicators (TI) as part of market analysis, whereas computer-assisted modelling includes Machine Learning (ML) for market analysis. This paper presents an ML-based stock market forecasting model in which Technical Indicators (TI) along with historical stock data are used to forecast the stock market movement. Tree-based ensemble models are trending due to their robust performance. XGBoost is one such tree-based ensemble model. However, hyperparameters of XGBoost are quite difficult to select. Therefore, an automatic selection of hyperparameters is required. This paper proposes optimized XGBoost in which popularly used optimization methods are utilized for hyperparameter optimization. For hyperparameter tuning, Grid Search (GS), Random Search (RS), Bayesian Optimization (BO), and Particle Swarm Optimization (PSO) are used. The performance of the hyperparameter-tuned XGBoost is tested on the three stock indices i.e. NIFTY, NASDAQ, and NYSE. The standard performance metrics Accuracy (ACC), Precision (P), Recall (R), and F1-Score (F1) are used for the model evaluation. Empirical studies show that PSO-optimized XGBoost provides the highest performance whereas GS-optimized XGBoost provides the lowest performance.