Development of tomato maturity level prediction model based on portable visible spectrometer and machine learning

Abstract

A tomato is classified as a fruit, which level of maturity is determined by its color. Upon distribution, tomatoes require sorting based on their ripeness level. Generally making improvements done conventionally with the human eye. This method has the disadvantage that the results are subjective. One way that can be used to measure the ripeness level of tomatoes is using a spectroscopic sensor. Spectroscopic sensors can predict the level of ripeness and its contents automatically. This study uses machine learning to create a model to classify ripeness level and predict firmness, total dissolved solids (TDS), and total acid in tomatoes. This study used tomatoes with 3 categories of maturity. Tomatoes were tested non-destructively, namely measuring firmness, total dissolved solids content, and total acid. The data obtained were processed using the Partial Least Square Regression method to predict firmness, TDS, and total acid, while the maturity level used the Naive Bayes method. The data processing results to predict the level of maturity using Naive Bayes obtained a success rate of 100%. While for the predictions of firmness, TDS, and total acid had R2 training and R2 testing, namely 0.685 and 0.678, 0.534 and 0.521, and 0.352 and 0.349, respectively.