EVALUATING CLASSIFICATION ALGORITHMS FOR PREDICTING SOCIAL PROJECT APPROVAL IN POST-CONFLICT REGIONS

OpenAlex Turkish

Implementing Social Projects (SPs) has become crucial in crisis-affected areas for supporting disadvantaged groups and decreasing poverty. The use of classification algorithms to forecast social project selection outcomes in post-conflict areas is examined in this study. Nine project variables, including financial, technical, spatial, and social aspects, were employed as predictive features based on a dataset that included 274 possible projects in Northern Syria. Logistic Regression, Simple Logistic, Naive Bayes, IBk (k-Nearest Neighbors), J48 decision tree, and Multilayer Perceptron were the six classification techniques that were assessed. Model performance was evaluated using accuracy, the Kappa statistic, mean absolute error (MAE), and root mean square error (RMSE). The results show that Simple Logistic and Naive Bayes obtained the best accuracy (98.18%) and Kappa (0.963), and Logistic Regression had the lowest MAE. The novelty of this study lies in using a real archival non-governmental organization (NGO) dataset from northern Syria to develop a machine learning-based decision-support framework for project-level social project selection, thereby providing a practical complement to traditional multi-criteria decision-making MCDM methods.