Generative AI-based Decision Support System for Detecting Palm Fruit Quality

Abstract

Indonesia's dominance as a global palm oil producer is challenged by the economic precarity of its 6.9 million smallholder farmers,dual threat to their sovereignty. The first is a structural asymmetry of information and weak bargaining power, making them vulnerable to subjective assessments of Fresh Fruit Bunch (FFB) quality, leading to unilateral price cuts and significant income losses. The second is mounting global market pressure to comply with sustainability standards (ISPO/RSPO),which act as non-tariff barriers through complex traceability and documented quality controls that are difficult for smallholders to meet. To address these challenges, this study designed and evaluated a generative AI-based Decision Support System (DSS) to provide objective analysis and strengthen farmers' decision-making capabilities. The methodology involved using few-shot prompting on the Gemini AI platform, with the system's performance then tested through a controlled technology intervention involving 20 smallholder farmers in South Lampung, a region with a high incidence of price disputes. Key findings indicate that the DSS, through real-time image analysis, can classify FFB maturity with a quantitative accuracy of 80%. Its implementation in the field consistently resulted in a 45% reduction in price disputes and an average increase in selling prices of 9-13%, as farmers utilized validation data from the system during negotiations. The technology achieved a high adoption rate of 80%, driven by increased farmer confidence and direct positive impacts on their income. In conclusion, this DSS has proven not only a valid technical tool but a crucial catalyst for economic empowerment, effectively bridging the information gap and empowering farmers for integration into a more equitable and sustainable global supply chain. Further research is recommended to test the system's scalability and generalization across larger populations in regions like North Sumatra or Kalimantan to validate its efficacy amidst diverse agronomic and socio-economic characteristics