Modeling rice productivity clustering with random forest: implications of regency agribusiness in 1986–2023

Rice productivity in Indonesia is strongly influenced by variations in environmental conditions, land management, and resource availability, creating disparities between high- and low-productivity areas. This study aims to segment regions based on rice productivity using data-driven clustering analysis to identify key patterns and influencing factors. A descriptive quantitative design was employed, applying Random Forest Clustering to annual rice productivity data (1986–2023) from 29 districts in Central Java, Indonesia, sourced from the Ministry of Agriculture. Data preprocessing, clustering, and visualization were conducted using JASP software. Model optimization used the Bayesian Information Criterion (BIC), and performance was evaluated via the Silhouette Score, Dunn Index, and Calinski-Harabasz Index. Three clusters emerged: high (mean = 6.8 quintals/ha), medium (4.5 quintals/ha), and low (2.9 quintals/ha). The model showed a Dunn Index of 0.396 and Calinski-Harabasz Index of 10.088, with Silhouette Scores ranging from 0.143 to 0.207, indicating moderate cluster separation. Results reveal a strong association between rice productivity and land management, environmental conditions, and agricultural inputs. This data-driven approach enables targeted interventions and supports evidence-based agribusiness strategies to optimize rice production in Indonesia.