Predictive Price-Signal Forecasting Using Machine Learning to Support Agricultural Supply Planning in Nigeria

An estimated 38 million tonnes of agricultural produce in Nigeria are lost to waste annually, largely due to a lack of reliable market intelligence for farmers before harvest. This paper presents a machine learning system designed to address this information gap by forecasting commodity price signals, used as a proxy for market activity, across 39 agricultural commodities. Using price and climate data from the World Food Programme, we evaluated seven algorithms and three conventional baselines through a structured tournament. Our results show that Random Forest outperformed other models for 19 of the 39 commodities (48.7%). Feature importance analysis reveals that recent historical price trends are significantly more predictive of future market signals than climate or seasonal variables. The system was deployed as a lightweight Flask web application that provides a market activity index to assist in harvest timing and distribution planning. We conclude with a discussion of the ethical implications of using price proxies for decision support and the importance of addressing representational bias in agricultural data.