Early Warning System for Local Government Financial Distress Using Support Vector Machine with Synthetic Tabular Data

Financial distress in local governments occurs when a region is unable to fulfill its debt obligations, creating risks to fiscal stability and regional economic health. This condition is especially concerning for governments that rely heavily on loans to fund infrastructure projects, as repayment failures can lead to penalties, additional costs, and limited access to future credit. Detecting the financial distress early is therefore crucial. However, prediction is often challenged by the imbalance data, out of 508 local governments in 2022, only 93 were classified as financially distressed, while 415 were financially healthy, resulting in a 1:4 ratio. Conventional predictive models tend to perform well for the majority class but fail to recognize high risk cases. To overcome this, the study applies synthetic data generation using the Tabular Variational Autoencoder (TVAE) and Conditional Tabular Generative Adversarial Network (CTGAN) to balance the dataset through oversampling. Support Vector Machines (SVM) with different kernels (linear, RBF, sigmoid, and polynomial) were trained on the augmented data. Model performance was evaluated using confusion matrices, F1-Score, G-Mean, and ROC-AUC. The findings indicate that SVM using the linear kernel trained with TVAE achieved superior results, with a G-Mean of 0.74 and ROC-AUC of 0.83, outperforming CTGAN based models (0.73 and 0.77). Feature importance analysis identified Individual Risk, Audit Opinions, and Human Development Index as key predictors of financial distress. Overall, integrating synthetic oversampling with SVM offers a practical and effective framework for developing early warning systems to help policymakers detect and mitigate fiscal risk.