Forecasting Rice Production in Mozambique Using Time Series Models and Artificial Neural Networks: Implications for Food and Nutritional Security in the Context of SDG 2

Abstract

Filipe Mahaluca, Faizal Carsane and Alfeu Vilanculos

The study presented a comparative analysis between ARIMA models and LSTM neural networks for forecasting rice production in Mozambique, covering the period from 2023 to 2030. Initially, the ARIMA(1,1,0) model was identified and fitted based on ACF and PACF plot analysis, followed by parameter estimation using the maximum likelihood method. Validation was conducted through diagnostic tests applied to the residuals, such as the Box-Pierce and ARCH tests, and model performance was evaluated using metrics like AIC, BIC, HQIC, RMSE, and MAPE. Concurrently, the LSTM neural network was configured with two LSTM layers of 50 units, trained with normalized historical data from 1961 to 2013, and validated with data from 2014 to 2022. To enhance the robustness of the forecasts, the Bootstrapping technique was applied, generating multiple data samples to calculate 95% confidence intervals. The results showed that the LSTM model outperformed the ARIMA(1,1,0) in terms of accuracy, with an average MAPE of 5.58%, compared to ARIMA's MAPE of 7.99%. Both models indicated a trend of stabilization in rice production over the years, suggesting a possible maturity stage in the country’s agriculture. However, the LSTM model's superiority in capturing nonlinear patterns and long-term dependencies makes it the more suitable model for future projections. These forecasts are crucial in the context of Mozambique's food and nutritional security, as they underscore the urgent need for strategic interventions and investments in agricultural technology to stimulate production growth. Addressing these challenges is vital for achieving Sustainable Development Goal 2 (SDG 2), which aims to end hunger and ensure food security by 2030. The findings provide a solid foundation for agricultural planning and the formulation of effective public policies to support food and nutritional security in Mozambique.

PDF