Ranking Cocoa Land Suitability Growth Requirements with Random Forest for a Spatial Decision-Support Platform

Abstract

Cocoa-growing land in Indonesia has steadily declined, increasing import dependence. An earlier study developed a web-based geographic information system (GIS) platform using Spatial ID3 to classify cocoa land suitability in Bogor Regency from eight soil and topographic variables, but it could not identify which variables most strongly influenced the resulting classes. This study extends it by using Random Forest to rank the eight variables by their contribution to the suitability decision. Trained on 238 field-survey records with class-imbalance handling, the model reached 0.89 accuracy and 0.80 macro-F1 under stratified five-fold cross-validation. Partial dependence analysis identifies, for each variable, the level most associated with the moderately suitable (S2) class, the nearest attainable target since no land reached the highest class. Results are overlaid on the nature reserve, protected forest, and karst zones of the 2024–2044 regency spatial plan to flag legally restricted land. Relief/slope (17.91%), soil pH (16.33%), and soil mineral depth (14.96%) were most influential, jointly accounting for 49.20% of total importance, while drainage contributed least (6.90%). The findings are delivered through a multi-layer suitability map, a growth-requirement priority panel, and a technical recommendation module, retaining the validated Spatial ID3 classification. Combining interpretable classification, data-driven factor prioritization, and regulatory compliance gives land-management decisions a clearer, actionable, and legally sound basis, supporting Indonesia's cocoa self-sufficiency.