Model Machine Learning untuk Klasifikasi Mutu Telur Ayam Ras Berdasarkan Kebersihan Kerabang

Abstract

Eggs are one of the food needs that have high enough nutrition and are the main needs for the community. Therefore, chicken eggs consumed by the community must have good quality so that they are safe and useful when consumed by the community. According to SNI 3926:2008, the quality of eggs is divided into 3, namely quality I, quality II and quality III in terms of external and internal conditions. In this study, a classification was carried out to determine the quality of chicken eggs based on the cleanliness of the shell using a machine learning approach. Several methods in machine learning such as KNN, Naïve Bayes, Decision Tree and SVM are used to classify chicken egg quality based on shell cleanliness. The stages in this research include data acquisition, preprocessing, feature extraction, modeling and model evaluation. A total of 90 chicken egg image data were used in this study with a total of 30 images in each of each quality. In the preprocessing stage, filtering and conversion of images to grayscle images is carried out. To get the texture features of the chicken egg image, feature extraction is carried out using first and second order statistical calculations. The results of feature extraction obtained mean, skewness variance, kurtosis, entropy, Angular Second Moment (ASM), contrast, correlation variance, inverse different moment, entropy. These feature values are then partitioned into training data and test data. A total of 3 experiments were carried out to perform classification, namely the composition of the comparison of training data and test data. Three experiments were carried out by varying the ratio of the amount of training data and test data as much as 70%: 30%; 75%:25% and 80%:20%. The next stage is classification using several methods in machine learning, namely KNN, Naive Bayes, Decision Tree and SVM. The method that produces the best accuracy is KNN and Decision Tree with an accuracy of 96% with a comparison composition of training data and test data of 70%: 30% and 75%: 25%.