Modeling and Forecasting Bean Production in Mozambique: Challenges and Implications for Food Security and SDG 2

This study examines the effectiveness of ARIMA and LSTM models in forecasting bean production in Mozambique, using data from 2002 to 2022. The analysis reveals that the limited sample size, comprising only 21 years of data, significantly impacts the accuracy of both models, as reflected in high MAPE values. The ARIMA(1,1,1) model demonstrates robustness with the lowest RMSE among the ARIMA models, but the LSTM model, despite its challenges, shows superior capability in capturing nonlinear patterns, resulting in a lower average MAPE. Forecasts for the period from 2023 to 2030 suggest stable bean production with slight annual variations, although the wide confidence intervals highlight the inherent uncertainty in these predictions. This study underscores the importance of improving forecasting models to better guide agricultural planning and policy-making, particularly in the context of Mozambique's food insecurity challenges and the global objectives of SDG 2. The results emphasize the need for more extensive data collection and the inclusion of additional variables to enhance the accuracy of future forecasts, contributing to the reduction of food insecurity and the achievement of sustainable development goals in Mozambique.