Performance Evaluation of Self-Organizing Feature Maps and Support Vector Machines in Predicting Stock Prices: A Comparative Study

Stock market forecasting plays a critical role in guiding investors and policymakers in dynamic financial environments. Despite advancements in predictive modeling, the comparative evaluation of machine learning techniques, such as Self-Organizing Feature Maps (SOFM) and Support Vector Machines (SVM), within the Nigerian Exchange Group (NGX) context has been limited. This study addresses this gap by investigating the performance of SOFM and SVM in predicting stock prices for five NGX-listed companies: United Bank for Africa (UBA), First Bank of Nigeria Holdings (FBNH), Guaranty Trust Bank (GTB), Nestlé (NESTLE), and Dangote Cement (DANGCEM), spanning financial, consumer goods, and industrial sectors. The dataset consisted of approximately 2,665 daily stock price observations (about 533 per company), covering the period 2011–2019. Following data cleaning and quality checks, preprocessing included Min-Max normalization and transformation into time-series matrices to ensure robustness and consistency. The dataset was divided into training (2011–2016, about 70%) and testing (2017–2019, about 30%) periods. SOFM was utilized for clustering and pattern recognition, while SVM incorporated technical indicators such as moving averages and price fluctuations. Implementation was conducted in MATLAB R2018a with a custom graphical user interface (GUI) for result visualization. Results revealed that SVM consistently outperformed SOFM across all datasets. For the UBA dataset, SVM achieved superior metrics, including an accuracy of 90.63%, specificity of 88.89%, and F1-score of 92.82%, with a computation time of 33.10 seconds. In comparison, SOFM demonstrated slightly lower performance with an accuracy of 88.13%, specificity of 85.19%, and an F1-score of 90.91%, and a computation time of 47.02 seconds. These findings establish SVM as a reliable and efficient model for stock price prediction on the NGX. Future research could explore hybrid models and broader datasets to enhance predictive accuracy and applicability in real-time investment strategies and risk management.