<![CDATA[Optimizing agricultural productivity in marginal areas of Merauke Regency, South Papua Province, Indonesia, faces significant challenges due to a high yield gap and low input efficiency. This study proposes an innovative machine learning-based approach to evaluate and map the performance of upland rice farmer groups by using PFPL (Prospective Farmer Prospective Location) data, which only has previously been used administratively. By integrating a predictive model (Random Forest Regressor), success classification, and K-Means Clustering, this study builds an adaptive and replicable analytical framework to support Data-Driven agricultural decision-making. The analyzed dataset includes 30 farmer groups which containing technical information such as land area, seed use, pesticide use, and herbicide use, as well as actual and targeted yields. The feature engineering process yielded the input efficiency ratio as the primary variable. The Random Forest regression model achieved a near-perfect fit on the available dataset (R² = 0.95; RMSE = 0.41). However, given the limited sample size (30 farmer groups), the result should be interpreted cautiously and regarded as exploratory rather than conclusive. Cluster analysis revealed two segments: a high-input but inefficient group and an efficient group with very high yields. These results highlight that input quantity does not guarantee productivity without efficient use. This study not only expands the literature on agricultural intelligence but also offers a practical approach for policymakers to design efficiency-based interventions, incentives, and training. This approach is also relevant for accelerating digital transformation and food security in underdeveloped regions.]]>
Copyrights © 2026
