Design of an expert system for sorting pistachio nuts through decision tree and fuzzy logic classifier

• Published in: Expert systems with applications Vol. 38; no. 4; pp. 4339 - 4347
• Main Author: Omid, Mahmoud
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.04.2011
• Subjects: Classification; Decision tree; Expert system; Fuzzy classifier; J48 algorithm; Pistachio nuts; Sorting; Fuzzy classifier; Expert system; J48 algorithm; Classification; Pistachio nuts; Decision tree; Sorting
• ISSN: 0957-4174; 1873-6793
• DOI: 10.1016/j.eswa.2010.09.103

► Design of expert system for sorting pistachio nuts. ► Novel use of J48 decision tree algorithm as feature selection and classification. ► Development of fuzzy logic classifier and its Simulink model for sorting pistachio nuts. ► New ideas for combining decision tree, fuzzy classifier system and Simulink in detection, classification and food quality inspection domains. This paper presents an expert system for sorting open and closed shell pistachio nuts. A prototype was set up to detect closed shell pistachio nuts by dropping them onto a steel plate and recording the acoustic signals that was generated when a kernel hit the plate. To determine the important characteristics and to unravel the significance of these signals, further analysis or processing was required. J48 decision tree (DT) is used for both feature selection and classification. Initially, the J48 DT was used for selecting the best statistical features that will discriminate among two classes from impact acoustic signals. The output of J48 DT algorithm was then converted into crisp IF-THEN rules and membership function sets of the fuzzy classifier. Four IF-THEN rules, generated from the extracted features of J48 DT, were required by the fuzzy classifier. To evaluate the performance of the expert system, data on 300 nuts of open and closed shells were used. The data were initially divided into two parts: 210 instances (70%) for training and the remaining 90 instances (30%) for testing the classifier. The correct classification rate and RMSE for the training set were 99.52% and 0.07, and for the test set were 95.56% and 0.21, respectively. These encouraging results as well as the robustness of the FIS based expert system makes the approach ideal for automated inspection systems.