Aplicación de machine learning para la orientación en la toma de decisiones frente al uso agrícola apropiado del suelo para zonas con cultivos ilícitos en Colombia

Abstract

The eradication of crops for illicit use in Colombia is the closest thing to a chimera: according to figures from the Drug Observatory, the source that offers official numbers on this matter, in 2018, 59,977 hectares of coca were eradicated manually, a laudable effort but that, without a doubt, falls short compared to the 169,018 hectares of coca registered that same year in the country. Agriculture in Colombia plays an important role in the economy and employment, although there is a common difficulty among Colombian farmers, which is that they do not choose the right crop based on productivity, climatic conditions, biota and soil properties, therefore productivity and quality are affected. The main objective of the research was to guide decision-making in determining the best agricultural use of the land where there are illicit crops in Colombia, through the application of machine learning, where as the main product an orientation instrument was obtained for growers who want replace illicit crops based on soil properties, climatic and biotic conditions, and productivity. For the development of this, the Knowledge Discovery in Databases KDD process was applied. The study evidenced that the best automatic methods to predict the substitutable crop are those based on Random Forest and that the learning application is an important decision support tool for crop prediction.