Research On Error Correcting Output Coding Algorithm And Its Applications

Multiple classifier fusion system is a hot research topic in recent years in many data mining technology. It uses multiple classifiers to solve the problem and can significantly improve the system's generalization ability, can get higher classification accuracy and robustness than the individual classifier. Multiple classifier fusion technology has received much attention by many experts.Using this technology to solve classification especially for multi-class problem is more urgent.Error correcting output coding algorithm is a typical method, which can deal with the multi-class problem.It decomposes one multi-class problem into several binary classification problems by using multiple classifier fusion system, and has been used in many real fields.This paper discusses the characteristics and existing problems of the traditional error correcting output coding algorithm.We extend this algorithm from the coding process, integration strategy, semi-supervised learning, and dynamic data environment so as to broaden its application field.We design and carry out the extended algorithms. Major works include:(1) Semi-supervised based Hierarchical ECOC algorithm (Semi-HECOC). Use hierarchical coding method and semi-supervised learning technology to improve the data adaptability.(2) A Hierarchical ECOC Algorithm based on KNNModel (KNNM-HECOC). Extend the fusion method, and improve the capability of classifier.(3) A method for handling the problem of concept drifting (IKnnM-DHecoc). Adapt quickly to dynamic data environment and detect the concept drift problem effectifilly.(4) A Feature Subspace Based Detection Algorithm for Concept Drifting (FSDA). Extend the way to detect the concept drift problem.Our research on error correcting output coding algorithm is valid, which were testd on different public datasets and real datesets. Our works promote and expand the research of the error correcting output coding algorithm, and have certain application value.