<![CDATA[Disease-related yield losses represent a critical constraint to cocoa productivity in Indonesia, particularly in South Sulawesi Province where endemic infections cause 30-70% production declines annually. This study evaluated the economic viability of implementing machine learning-based disease detection systems in cocoa plantations in Luwu Regency through an 18-month mixed-methods research design integrating technical validation, randomized controlled field trials, and comprehensive economic analysis. A convolutional neural network model was developed using 12,450 labeled images and deployed across 30 cocoa farms stratified by size and disease pressure, with 15 treatment farms receiving ML-based detection technology and 15 control farms continuing conventional monitoring practices. The ML model achieved 93.7% diagnostic accuracy for detecting Cocoa Pod Borer, Vascular Streak Dieback, and Black Pod Disease. Treatment farms demonstrated significantly higher yields (1,247 kg/ha vs. 942 kg/ha, 32.4% increase), reduced disease incidence (8.7% vs. 23.1%), and improved bean quality (73.2% Grade A vs. 58.4%). Economic analysis revealed highly favorable investment returns with Internal Rate of Return of 47.3% for individual adoption and 52.6% for cooperative models, Net Present Value of $2,847 per farm, Benefit-Cost Ratio of 3.68, and Payback Period of 2.8 years. The findings demonstrate that ML-based disease detection achieves economic viability in smallholder cocoa farming contexts, offering a transformative solution for enhancing agricultural productivity and farmer incomes in disease-endemic tropical plantation systems. ]]>
Copyrights © 2025
