A Computational Intelligence Approach to Predicting Agricultural Commodity Price Volatility in Nigerian Markets

Agricultural commodity price volatility in Nigeria causes significant welfare consequences for smallholder farmers and urban consumers alike, yet the predictive modelling of this volatility using computational intelligence methods tailored to Nigerian market dynamics has not been adequately developed in the literature. This study develops and evaluates a suite of computational intelligence models for predicting the price volatility of four key agricultural commodities, namely maize, sorghum, groundnut, and yam, across Nigerian market zones covering all six geopolitical regions. The methodological contribution centres on the design of a novel Hybrid Recurrent Convolutional Neural Network architecture that integrates temporal price series data with climate, conflict, and transportation disruption covariates specific to Nigerian commodity market dynamics. The hybrid model is benchmarked against support vector regression, long short-term memory networks, gated recurrent units, and ensemble random forest methods using 15 years of daily price data from 24 markets sourced from the National Bureau of Statistics and the Federal Ministry of Agriculture. The proposed hybrid architecture achieves a mean absolute percentage error of 4.3 percent on a held-out test set, outperforming all benchmark models by a margin of at least 1.8 percentage points. Shapley additive explanations analysis identifies conflict proximity, rainfall deviation, and fuel price changes as the three most influential predictors of price volatility spikes. The model has direct applications for early warning systems operated by the Federal Government's price stabilisation programmes. Keywords: computational intelligence, commodity price prediction, neural networks, Nigerian agriculture, price volatility.