Artificial neural network modelling in the prediction of bananas' harvest

Banana tree (Musa spp.) is responsible for providing one of the most consumed and appreciated fruits in all regions of the world, and is cultivated mainly in tropical countries. In this connection, several management systems have been developed to simulate growth, yield, as well as the production of several crops according to climatic data. This study seeks to investigate the relationship of climatic variables in the banana bunch gestation period in order to predict the time of production. For that purpose, it was used an artificial neural network to estimate the bananas' harvest period in subtropical regions. The experiment was conducted for 7 cycles/years using Islanicao' cultivar. Climatological data were measured by automatic stations. According to the results' analysis, it can be verified that the estimation of the harvest through artificial neural networks presented 0.3% error and coefficient of determination of R-2 of 89%. From the developed model it was possible to establish the banana harvest forecast. It can be verified that the RNAs present a high percentage of correctness in the collection of the harvest, this is confirmed by the low square error. In this way, the model becomes a management tool for banana producers to help forecast demand.