Feature Selection based on Wrapper Method to Classify Purity of Palm Civet Coffee Shinta Widyaningtyas (a*), Muhammad Arwani (a), Sucipto (b), Yusuf Hendrawan (c), Ircham Ali (d)
(a) Department of Agroindustrial Techonology, Universitas Nahdlatul Ulama Indonesia, Jl.Taman Amir Hamzah 5, Jakarta Pusat, 10320, Indonesia
(b) Department of Agricultural and Biosystems Engineering, Universitas Brawijaya, Jl. Veteran, Malang, 65145, Indonesia
(c) Department of Agricultural Engineering, Universitas Brawijaya, Jl. Veteran, Malang, 65145, Indonesia
(d) Department of Informatics Engineering, Universitas Nahdlatul Ulama Indonesia, Jl.Taman Amir Hamzah 5, Jakarta Pusat, 10320, Indonesia
Abstract
Palm Civet Coffee is one of the specialty coffee and well known as the rarest coffee due to their limited production. This causes green bean palm civet coffee counterfeited with regular green bean coffee. However, the absence of tools and methods to classify Palm Civet Coffee coffee counterfeiting makes the sensing methods development urgent. Classify purity in Palm Civet Coffee can be utilize machine vision as nondestructive sensing. Machine vision can extract 101 image features consist 8 color features, 3 morphological features, and 90 textural features. The purpose of this study is identify the optimal subset features to classify purity of palm civet coffee. In this study, wrapper method feature selection based on K-Nearest Neighbours (KNN) Classifier with 13 algorithms used to select informative image features based on accuracy and f1-score. The result showed that Sine Cosine Algorithm had best performance as a feature selection technique with accuracy and f1-score at 0,975 with the selected features are S_HSV Contrast, Blue_Variance, Gray_Variance, S_HSV Variance.
