Artificial Intelligence’s Role in Predicting Corporate Financial Performance: Evidence from the MENA Region

This study classifies corporate financial performance in countries in the Middle East and North Africa (MENA) region, addressing the critical need for accurate and early identification of high-, moderate-, and low-performance companies. The selection of the MENA region was driven by its significant economic growth, diverse market structures, and increasing attractiveness for foreign investment, which makes accurate financial performance assessment important. Despite the growing interest in AI applications for corporate financial performance, a research gap still persists. Existing studies focus primarily on bankruptcy and financial distress prediction in developed countries, with rather limited studies on multi-class financial performance classification in the MENA region. This study addresses a significant gap in the corporate financial performance evaluation literature, which is the lack of a robust, comparative evaluation of advanced DL techniques against conventional ML methods for multi-class corporate financial performance prediction using high-dimensional data. This study employs a design science research (DSR) approach by developing an evaluation analytics artifact that integrates structured preprocessing, dimensionality reduction, and comparative ML and DL modeling, following the relevance, design, and rigor cycles. By employing a design science research (DSR) methodology, the research used a dataset from the Compustat database, comprising 7971 firm-year observations from 2013 to 2024. A rigorous dimensionality reduction process, including pairwise correlation filtering, resulted in a final set of 15 key classification features. The study compared three machine learning techniques—random forests (RFs), support vector machines (SVMs), and eXtreme Gradient Boosting (XGBoost), against one deep learning technique, deep neural networks (DNNs), for classifying the corporate financial performance of MENA-region companies. The models were trained to classify corporations into three performance classes (low, moderate, and high), using the earnings per share (EPS) as the target variable. The empirical findings indicate that all four machine learning algorithms achieved meaningful predictive performance in classifying EPS-based corporate performance. Among the benchmark models, the support vector machine (SVM) and random forest (RF) classifiers produced stable and competitive results, indicating strong generalization capabilities across firms and periods. XGBoost consistently outperformed the traditional machine learning models, delivering the highest overall classification accuracy and superior discriminatory power, highlighting its effectiveness in capturing nonlinear relationships and complex feature interactions. Similarly, the deep neural network further improved classification performance relative to the benchmark models and exhibited comparable results to XGBoost, especially in modeling high-dimensional data. This superior performance can substantially enhance earnings performance classification through early performance deterioration and improvement identification, allowing more proactive strategic and operational decisions.