Using GEP for Rule-Based Classifier

• Main Authors: Lin-Cheng Chen, 陳玲赬
• Other Authors: 林志麟
• Format: Others
• Language: zh-TW
• Published: 2008
• Online Access: http://ndltd.ncl.edu.tw/handle/64520910337270619363

碩士 === 元智大學 === 資訊管理學系 === 96 === Building of classification rules is not only to bestead prediction of instance in the future but also to understand characteristics of each kind of data. This study proposes a rule-based classifier method based on Gene Expression Programming. The proposed method uses Automatically Defined Functions to build important substructures efficiently, and consequently produces chromosomes of high quality. According to different purposes of the experiment, there are two experiments. In the experiment 1, it focus on small domain of attribute in three data sets(WBC、Dermatology and Mamographic Mass Data). In the experiment 2, it focus on large domain of attribute in two data sets(WDBC and Diabetes). Moreover, according to our experimental results, the prediction efficiency of the proposed method is better than GEPCLASS, in two data sets(WBC and Dermatology). The proposed method is compared with WEKA (Naïve Bayes、Decision Table、J48 and Logistic), using three data sets (Mamographic Mass Data、WDBC and Diabetes). According to our experimental results, the prediction efficiency of the proposed method is better than WEKA.