Atatürk Üniv. Ziraat Fak. Derg., 49 (1): 37-43 , 2018 Atatürk Univ., J. of the Agricultural Faculty, 49 (1): 37-43 , 2018 ISSN : 1300-9036 Araştırma Makalesi/Research Article Data Mining Approach For Prediction Of Fruit Color Properties Bünyamin DEMİR 1* Feyza GÜRBÜZ 2 İkbal ESKI 3 Zeynel Abidin KUŞ 4 1/ Mersin University, Vocational School of Technical Sciences, Department of Mechanical and Metal Technologies, Mersin, Turkey 2/: Erciyes University, Faculty of Engineering, Department of Industrial Engineering, 38039, Kayseri, Turkey 3/: Erciyes University, Faculty of Engineering, Department of Mechatronics Engineering, 38039, Kayseri, Turkey 4/: Erciyes University, Faculty of Agriculture, Department of Biosystems Engineering, 38039, Kayseri, Turkey (*Corresponding author email: bd@mersin.edu.tr) Geliş Tarihi :13.12.2017 Kabul Tarihi :08.01.2018 ABSTRACT: Color is an important feature that dictates the quality and consumer preferences of many fresh fruits and vegetables. In color measurement of fruits, the CIE L*a*b* color space is widely used since it is a uniform color scale. In this study, raw data for the color features of apple varieties were divided into two parts as test and train data in the first stage, analyses were performed on train data and tests were performed on test data. The rules obtained by applying the Find laws algorithm were used to estimate the color index (CI), hue angle (h *) and Chroma (C *) values. In the second stage, raw data were classified by Strict and Liberal options of cluster analysis. Find Laws algorithm was applied to each cluster and 7 different prediction rules were obtained for CI, h*and C* parameters. R 2 values of the rules were compared and the rules with the most accurate outcomes were identified. Keywords: Apple, hue angle, L*a*b*, color space. Meyve Renk Özelliklerini Tahmin Etmek İçin Veri Madenciliği Yaklaşımı ÖZET: Renk, birçok taze meyve ve sebzenin kalitesini ve tüketici tercihlerini belirleyen önemli bir özelliktir. Meyvelerin renk ölçümünde, uniform renk ölçeği nedeniyle CIE L*a*b* en çok kullanılan renk uzayıdır. Bu çalışmada elma çeşitlerinin renk özelliklerine ait ham veriler ilk aşamada test ve eğitim verileri olarak iki kısma ayrılmış, eğitim verileri üzerinde analizler yapılmış ve test verileri ise testlerde kullanılmıştır. Find laws algoritması uygulanarak elde edilen kurallar Color index (CI), hue angle (h*) and Chroma (C*) değerlerini tahmin etmek için kullanılmıştır. İkinci aşamada ise ham veriler cluster analizine tabi tutularak Strict ve Liberal seçenekleri ile sınıflandırılmıştır. Find laws algoritması her bir sınıfa tek tek uygulanıp, her bir CI, h*, C* parametreleri için elde edilen 7 farklı tahmin kuralı R 2 değerlerine göre karşılaştırılarak en yüksek doğruluğa sahip kurallar tespit edilmiştir. Anahtar kelimeler: Elma, hue açısı, L*a*b*, renk uzayı INTRODUCTION Visual appearance is the primary quality attribute for foodstuffs and it is also the first quality aspect looked for by the consumers (Maskan, 2001). For fruits, size, shape, texture, color and surface defects are the basic external quality attributes. All these attributes are also related to visual appearance of the fruits and agricultural products (Zhang et al., 2017). Among these visual quality attributes, color is the most significant parameters used as an indicator of the quality. Therefore, majority of the consumers first look at the color of fruits, vegetables and meats to judge the quality of those foodstuffs (Wu and Sun, 2013; Trinderup at al., 2015). Peel color is the key quality attribute for apples. It not only influences consumer preferences, but also is related to nutritional values of the apples. Peel color is also used to distinguish one cultivar from the other since each cultivar has specific color characteristics (Rabinovich, 2009). Multiple color spaces are often used to define color parameters of fruits. Among them, CIE L*a*b* (CIELAB) specified by International Commission on Illumination is the most common one (Fairchild, 2013). Data mining approach uses various tools and techniques to inquire meaningful data from a large data set. It has recently started to be used in agricultural researches and implementations (Chowdhury and Ojha, 2017). Several previous researches employed data mining techniques in agricultural researches to predict the value of an attribute by using already available measurements of that attribute (Ramesh and Vardhan, 2013; Gonzalez- Sanchez et al., 2014; Veenadhari et al., 2014; Pantazi et al., 2016; Germšek et al., 2017; Isaza et al., 2017; Kus et al., 2017; Majumdar et al., 2017). Data mining approach includes four basic methodologies as of clustering, classification, feature selection and outlier detection. On the other hand, artificial neural networks, decision trees, k-means type algorithms, genetic algorithms, nearest neighbor method and rule induction are the primary techniques