Explainable machine learning for predicting the success of IT projects in Nigerien SMEs: an empirical study based on 488 projects in Niamey

IT projects play a crucial role in the digital transformation of small and medium-sized enterprises (SMEs), but their success remains highly dependent on organizational, financial, and infrastructural constraints, particularly in developing countries. This study proposes an explainable predictive framework for the success of IT projects applied to SMEs in Niamey, Niger's main digital hub. The analysis is based on a set of 488 IT projects and compares the performance of seven machine learning algorithms: logistic regression, decision tree analysis, random forest analysis, support vector machines (SVMs), k-nearest neighbors, gradient boosting, and XGBoost. A rigorous methodology integrating data preprocessing, hyperparameter optimization using GridSearchCV, and cross-project validation based on GroupKFold was implemented. The results show that the SVM model achieved the highest predictive performance with an AUC-ROC of 0.850. To improve the interpretability of the predictions, explainable artificial intelligence analyses based on SHAP were conducted. These analyses revealed that stakeholder satisfaction, budget overruns, delays, project complexity, and certain constraints related to power supply and internet availability are the main determinants of project success. The study also led to the development of a prototype early warning system designed to support decision-making and proactively manage IT projects in SMEs.