การเปรียบเทียบวิธีการวิเคราะห์จำแนกประเภทกับการวิเคราะห์ความถดถอยโลจิสติกอันดับสำหรับการจำแนกกลุ่มแบบอันดับ / วิศวัฒน์ รสหวาน = A comparison of discriminant analysis and ordinal logistic regression for ordered-group classification / Visawat Roswarn
The purpose of this research is to study and compare the accuracy of classification for ordered-group between Discriminant Analysis (DA) and Ordinal Logistic Regression (OLR). In this study, we focus on multivariate normal independent variables which cover independent case and dependent cases; coefficient of simple correlation = 0.1, 0.5, 0.9 . We also compared the assumption of homogeneity of covariance matrices among groups. The numbers of groups are 3, 4 and 5 groups in which the 120, 240 and 360 samples were simulated equally, moderately-equally and unequally. The data for this research was simulated by using the Monte Carlo simulation technique with 1,000 repetitions for each case. The criterion of determination is Apparent Rate Error (APER). The results of this research are Ordinal Logistic Regression classifies ordered-groups more accurately than Discriminant does under every condition and the number of ordered group has the most effect on accuracy of classification while the equality of sample in each group has the least effect on the accuracy.