Comparing Deep Learning and Traditional Machine Learning for Predicting Governance Outcomes: A Comprehensive Benchmark

Predicting governance outcomes has become increasingly important for improving public sector performance, allocating resources, and identifying municipalities at risk of underperformance. While traditional machine learning (ML) models such as Random Forest (RF), XGBoost, and Multiple Linear Regression (MLR) have been widely used in policy analytics [1], [2], recent advances in deep learning (DL), particularly Convolutional Neural Networks (CNNs) and Multilayer Perceptrons (MLPs), offer new opportunities for capturing non-linear, high-dimensional patterns in governance data [8], [10]. This study conducts the first comprehensive benchmark comparing DL and ML approaches for predicting local governance performance across a national dataset of municipalities. Results show that MLP models outperform all traditional models, while CNNs achieve competitive performance but require more data and computational resources. The findings highlight the conditions under which each family of models is most appropriate, providing methodological guidance for data-driven public management [9].