A novel technique for Multi-class ordinal regression-APDC

Abstract

Objectives: Ordinal regression is one which is used in Multiclass classification where there is an essential ordering among the classes. The training dataset is initially classified depending on the Random threshold values?. Based on these values, the distance between the different class labels are predicted by one against one technique. Method: All Pairs Distance Calculation using one against one technique [APDC_1 AG 1] is Proposed to validate the work. But in the referred previous work, distance is calculated using adjacent classes, but here all pairs distance calculation is used to find the class label distance to all class label pairs. Findings: On the whole, New trained data are in the form of one dimensional representation. Here, with the knowledge of proposed work, testing data is tested with New trained data set and the results are produced. The Proposed method is seen to be ambitious when compared with previous work. Beside this, an additional set of experiments is done to study the potential quantifiability and illustratability of the proposed method when using APDC as base methodology. Improvements: Proposed work is analyzed with Kernel discriminant analysis, Logistic Regression, Classification via Regression, Multiclass Classifier and found APDC has attained better results according to all measures.