Analyzing Machine Learning Algorithms For Predictive Modeling Of Financial Securities: The Case Of Al_Rjehei Stock

Abstract

Predictability of financial markets is the cornerstone of computational finance and has important implications for healthy economic growth and security of investments. This research studies the performance of machine learning (ML) models on the closing prices for the AL_Rjehei stock using a long-term historical dataset which was from 2010 to late 2025. We analyze five different modeling methods: Random Forest, Gradient Boosting, Decision Trees, Ensemble, and Neural Networks. Using a strict data partitioning approach (Train, Validate and Test), we evaluated model efficiency using Average Squared Error (ASE) and Root Mean Absolute Error (RMAE). By experiment results we showed that the Random Forest was the "Champion model", with the lowest Average Squared Error of 0.9788 on test partition. A feature importance analysis revealed that the 'Open' price was the most important predictor, outperforming trading "Volume" in predictive weight quite a lot. Though Gradient Boosting performed competitively (ASE: 1.0117), the most error rate was found for the Neural Network architecture, which indicates a greater challenge in hyperparameter tuning or larger feature space for the deep learning task. The results reveal good stability and accuracy of ensemble bagging techniques for financial time-series data in terms of volatility. This study adds to the literature with a validated stock price prediction model to help institutional and individual investors make sound decisions in their search for economic health.