A Hybrid ARIMA-LSTM Model for Forecasting Agricultural Yield in South-Western Nigeria

Accurate forecasting of agricultural yield is crucial for economic planning and food security in developing economies. This paper proposes a hybrid forecasting model combining the linear statistical capability of Auto-Regressive Integrated Moving Average (ARIMA) and the non-linear high-dimensional capture capacity of Long Short-Term Memory (LSTM) recurrent neural networks. Using over four decades of crop production and rainfall records from southwestern Nigeria, we construct and validate the models. The empirical results demonstrate that the proposed ARIMA-LSTM hybrid model significantly outperforms individual ARIMA and LSTM approaches, reducing the Root Mean Squared Error (RMSE) by 24.5% and the Mean Absolute Percentage Error (MAPE) to under 4.2%. Furthermore, we examine the influence of macroclimatic anomalies on crop output. The findings suggest that blending classical statistical theories with modern deep learning architectures offers higher resilience and predictive stability for planning policies.