Evaluation of Long Short-Term Memory and Hidden Markov Model in Predicting the Productivity of Maize in Nigeria

Abstract

Background: Due to population increase and import constraints, maize, a key cereal crop in Africa, is experiencing a boom in demand. Appropriate models for predicting the productivity of the commodity to meet rising needs is a need of the hour. Objectives: The study determined the interactions between maize production in Nigeria and other climatic factors, mostly rainfall and temperature, and predict its outputs using models that encapsulate the relationship. Methods: The Hidden Markov Model (HMM) and the Long Short-Term Memory neural network (LSTM) are both evaluated and compared in this context to assess their performance in the prediction of the performance of Maize in Nigeria. A variety of performance indicators, such as correlation, mean absolute percentage error (MAPE), standard error of the mean (SEM), and mean square error (MSE), are used to evaluate the models. Results: The outcomes show that the HMM performs ten times better than the LSTM, with an RMSE of 1.21 and a MAPE of 12.98 demonstrating greater performance. Based on this result, the HMM is then used to forecast maize yield while taking the effects of temperature and rainfall into account. Conclusion: The estimates highlight the possibility for increasing local output by demonstrating a favorable environment for maize planting in Nigeria. In order to help the Nigerian government in its efforts to increase maize production domestically, these studies offer useful insights.