The use of Extreme Gradient Boosting in predicting stock prices of selected companies in the Iraqi Stock Exchange

Abstract

This paper analyzes stock price data from 20 companies listed on the Iraqi Securities Exchange over the period from early 2020 to late 2023, employing the Extreme Gradient Boosting algorithm to uncover optimal adjustable parameters and evaluation metrics for various stock symbols. Key parameters such as Max Depth, Gamma, Learning Rate, and N Estimators are examined to understand their influence on model complexity and prediction accuracy. The findings illustrate that a max depth of 12 or 18 provides a balance between capturing complex relationships and avoiding overfitting, with companies exhibiting varying depths corresponding to their unique performance drivers. Gamma values highlight the trade-off between model complexity and interpretability, where lower values allow for intricate relationships while higher values promote simplicity. Learning rates around 0.05 indicate stability in convergence, while varying N Estimators (400 vs. 500) affect model performance and training efficiency. The evaluation metrics, including MSE, RMSE, MAE, MAPE, and R², provide insights into the effectiveness of the models, with companies like Mosul Bank demonstrating low error metrics and strong predictive capabilities. Overall, the study emphasizes the importance of parameter selection and evaluation metrics in enhancing the prediction accuracy of stock price dynamics in the Iraqi market, offering valuable insights for investors and stakeholders.