<![CDATA[This study examines the level and dynamics of regional fiscal autonomy in Mamuju Regency, Indonesia, from 2013 to 2023 by integrating conventional public finance indicators with a graph-based machine learning approach. The primary objective is to measure fiscal independence and to evaluate the effectiveness of a Similarity-Based Multi-View Graph Convolutional Network (SMGCN) in modeling complex relationships among local revenue, intergovernmental transfers, and total regional income. Fiscal autonomy is operationalized using a fiscal independence index derived from the ratio of Locally Generated Revenue to total regional revenue, while additional variables capture transfer dependence and structural fiscal composition. The empirical analysis employs annual regional budget realization data, which are transformed into multiple graph views reflecting different fiscal dimensions. These graphs are then processed using SMGCN to classify and predict fiscal autonomy categories. The results indicate that Mamuju Regency exhibits persistently low fiscal autonomy, characterized by a high dependence on central government transfers and a limited contribution of Locally Generated Revenue. From a methodological perspective, the proposed model demonstrates superior predictive performance compared with conventional statistical and machine learning benchmarks, suggesting that graph-based representation learning is suitable for capturing interdependencies among fiscal variables. The findings contribute to the literature on fiscal decentralization by providing both empirical evidence of structural fiscal dependence at the local level and a novel analytical framework for assessing regional fiscal performance. Policy implications emphasize the need for strengthening local revenue bases and diversifying regional economic activities to enhance long term fiscal sustainability.]]>
Copyrights © 2026
