Modelo predictivo de producción agrícola colombiana utilizando redes neuronales y sistemas neuro difusos

Abstract

This work aims to get predictiv models of colombian agriculutral prediction use artificial neural networks and neuro fuzzy systems. Historical data were used for the production of six agricultural commodities. Data imputation and normalization were performed for their preprocessing, with which the models that use learning techniques and fuzzy logic were trained. Different input-output configurations were proposed for the models, in order to determine the best one from the value obtained from the calculation of the mean square error (MSE) for validation data. For neural networks, the best model is the one that considers the previous output value (production), and for neuro-fuzzy systems, the best model, both linear and constant, is the one that has an input associated with the year in which the production measurement is made. This research provides a basis for agricultural planning, hand in hand with the use of artificial intelligence as a current and relevant technological tool.