Maturity Level Detection of Sugar Apple (Annona squamosa L.) Based On Physical And Chemical Properties at Room Temperature Storage Using Random Forest and K-Nearest Neighbor (k-NN) Agorithms

Abstract

The classification process on fruit is often a problem in the production process of agricultural
products, one of which is in classifying the level of ripeness of srikaya fruit, and there have been
many kinds of research on classifying fruit using one or two parameters using machine learning.
Physical and chemical properties such as aroma, moisture content, total dissolved solids, texture,
and weight loss indicate fruit maturity. This study aimed to determine the maturity level of srikaya
fruit based on physical and chemical properties using the k-Nearest Neighbor (k-NN) and Random
Forest algorithms, and to measure the algorithm’s accuracy. The method used in this research is
the k-NN and Random Forest, then its performance is measured using a confusion matrix.
Physical properties (weight loss and texture) and chemical properties (moisture content, total
dissolved solids and gas content) were observed. The results of detecting the maturity level of
sugar apple have been reached. From 8 test data, it can detect 1 damaged fruit using the k-NN
method, while the random forest from 8 test data can detect 2 raw fruit, 3 ripe fruit, and 1 damaged
fruit. The accuracy of the k-NN method is 12.5%, and the random forest is 75%. The performance
of the random forest is higher than the k-NN method based on the accuracy, precision, sensitivity,
and specificity results.