Predictive models for the detection of diseases in crops through supervised learning

Abstract

This paper proposes a methodology for the application of prediction algorithms for the development of diseases in crops using data mining techniques, which incorporate validation mechanisms based on data analysis requirements including verification of the method of selection and presentation of the results, as well as mechanisms of validation of the results based on metrics of quality of the information, which guarantee the effectiveness in the construction of the knowledge. The conditions for the establishment and proliferation of diseases are used as a case study in the analysis and contrasted with the favorable meteorological conditions for the different diseases, using methods that allow the collection of data for the prognosis of the disease. The models relate indicators of occurrence with meteorological data collected from the National Institute of Meteorology and Hydrology located in Querochaca Experimental Farm of the Technical University of Ambato whose geographical coordinates are: Latitude: −1.353543; Longitude: −78.617175. The data analysis techniques used were able to predict crop diseases in 78.80% with the J48 algorithm and 79.18% with the Logistic Regression algorithm based on data collected and analyzed from the meteorological station of the year 2015–2016, allowing the iterative search of correlations of consecutive day records, agro-climatic variables and biological variables. Our study is an initial proposal taking as parameters the temperature and humidity from previous works that qualify this line of research.