Predicción del índice de madurez del limón cultivado en el suroeste del ecuador a través de algoritmos de machine learning

Abstract

The production of lemon (Citrus Aurantifolia Swingle) in the southwest or throughout Ecuador, is a very important activity within the country's economy, the automation of the fruit classification process, according to its maturation state, helps save time and minimize errors in tasks involving people. The objective of this research is to predict the maturity index of the lemon, using Machine Learning classification algorithms, thus reducing the errors that can occur using human sight to classify this fruit, being more accurate because it is based on the physicochemical parameters of the lemon. The methodology used in the research is divided into two stages: the first stage was to obtain the physicochemical parameters through titratable acidity, soluble solids, density and pH, which will then be correlated by means of a color chart to determine their different states immature (IN), mature (M) and overripe (SM); and the second stage consists of applying the Machine Learning classification algorithms in which two data sets are processed, training and Test, therefore the algorithms K-nearest neighbor (KNN), Naive Bayes, Support Vector Machines (SVM), Decision Tree, Random Forest and Neural Networks with a Test accuracy of 1,0 can predict the ripening stages (IN, M, SM) of the lemon, except for the Logistic Regression algorithm that showed a Test accuracy of 0,33, making its rate error is greater than 50%. Once the analysis of the data was concluded, it was concluded that 6 of the 7 algorithms are very good for the classification of the different maturational stages, except for one, in which it presented a very low Test accuracy.