Research And Application Of Error Correct Output Code

Error correcting output coding(ECOC)is a framework of ensemble learning method for multi-class classification,which offers a better classification effect than a single classifier.After twenty years of its birth,ECOC has been applied to face recognition,traffic signal recognition,microarray data analysis and other fields.The error correcting output coding framework consists of two stages:coding and decoding.The researches in this field concerns of these two aspects,but the most of work focuses on the design of coding algorithm.In this dissertation,the coding and decoding of error correction output coding are explored respectively.In the aspect of coding,this dissertation introduces the importance and different types of coding algorithms.It also introduces several classical coding methods,based on which two new coding methods are proposed.(1)Self-adjusting error correct output coding algorithm(SA-ECOC)for microarray data.The algorithm designs coding matrix according to the unique characteristics of microarray data.This method first generates a coding matrix,and then iteratively adds columns.The algorithm also proposes a column selection standard "distance interval" to ensure that the newly added columns improve the classification results.SA-ECOC was used in the experiments based on cancer microarray data.Some popular ECOC algorithms are deployed for comparisons.The results showed that the accuracy of SA-ECOC classification was a little higher and more stable.(2)Customization coding method(CECOC)analyzes the characteristics of data samples,and designs characteristic coding matrix for different samples.The method is verified by UCI data sets and microarray data sets,and the experimental results show that the CECOC method is better and more stable than the matrix without custom coding.In the aspect of decoding,this dissertation introduces the functions and types of decoding methods,and also introduces several common decoding methods.The attenuated Euclidean distance decoding(AED)is analyzed,and two shortcomings are put forward.Then a new decoding method,the ratio Euclidean distance decoding(RED),is proposed.