Predicting Business Failure in Morocco: A Machine Learning-Based Strategy for Enhancing Risk Management

With rising business failures in emerging economies like Morocco, there is a need for more accurate prediction models. Traditional statistical methods often struggle with complex financial data and have been mainly used in Moroccan studies. This work aims to fill this gap by applying advanced machine learning techniques to improve bankruptcy prediction. To address this gap, we apply six machine learning algorithms, DT, XGBoost, RF, SVM, MLP, and K-NN to a dataset of 124 Moroccan manufacturing firms for two years, 2018 and 2017. Using financial ratios as predictors and k-fold cross-validation for evaluation, The study’s findings show that tree-based models, mainly RF, XGBoost, and DT, had the best performance and most consistent accuracy across both prediction horizons (t-1, t-2) in identifying early signs of financial distress, such as accounts receivable turnover, intangible assets ratio, and ratio of sales to assets, while the SVM, MLP, and K-NN had lower levels of reliability, particularly at t-2, and had lower levels of recall and greater uncertainty concerning their predictions. The present study demonstrates the effectiveness of advanced machine learning in predicting bankruptcy at an early stage in developing countries and emerging markets such as Morocco, and provides crucial information to banks and other stakeholders, who are interested in understanding the strengths and weaknesses of each model. This can further enhance their ability to manage risk with a better understanding of local economic realities and industry settings.