Feature Selection on Magelang Duck Egg Candling Image Using Variance Threshold Method

Magelang ducks are the leading poultry commodity in Central Java because they have the best egg quality among other local duck eggs. Egg candling is performed to separate fertile and infertile eggs, which is limited to the accuracy of human vision. Egg sorting is carried out to increase the productivity of Magelang duck eggs. This research aims at finding the best feature in distinguishing fertile and infertile eggs using the variance threshold method with the K-Nearest Neighbor (KNN) algorithm on the Magelang duck egg candling image. The dataset used is 86 images of Magelang duck eggs with training and test data of 70:30. The characteristics used are color, shape, and texture of the Grey Level Co-Occurrence Matrix (GLCM) with 18 initial features. The research begins with image acquisition, then image pre-processing, which includes cropping, resizing, and segmentation, followed by feature extraction. After normalizing the results of feature extraction, feature selection is then performed and classified using the KNN algorithm. The highest evaluation result is 92.31% with a classification time of 0.008 seconds at a value of k=5 with features: green, roundness, variance, and standard deviation.