An Intelligent Model Validation Method Based on ECOC SVM

Abstract

This paper develops an intelligent model validation method based on error correcting output coding support vector machine (ECOC SVM). The similarity analysis between simulation time series from computerized model and observed time series from real-world system is formulated as a multi-class classification problem. The ECOC framework, built on the basis of the error correcting principles of communication theory, decomposes the multi-class classification task as multiple binary classification problems. The SVM is used as the base classifier and a set of similarity measure methods is applied to extract the input features. Compared to conventional methods, the proposed validation method based on ECOC SVM incorporates multiple similarity measures to a comprehensive similarity measure and can learn to predict the credibility level from training samples. The application result reveals that the classification accuracy achieved 82%, which means the proposed method is promising for the similarity analysis of large datasets.