Multi-labeled Ensemble For Weak Labeling And Stable Algorithm

With the rapid development of modern various technologies, more and more application in real life are combined with multi-labeled datasets. Hence, the method and application of multi-labeled classification has become a hot topic for researchers in data mining and machine learning, among of which multi-labeled classification based on ensemble learning is one problem much worthy of being studied.Ensemble learning can improve the performance of classifier. Adaboost is the one of ensemble learning algorithm, but in previous it is studied based on unstable algorithm, for an example decision tree, neural network. This thesis expands the Adaboost algorithm, which describes it base on MLKNN and gives a specific method for base learners. This thesis presents Adaboost.ML algorithm, and the basic idea is that the base classifier of Adaboost is MLKNN and the method of Adaboost is changed in order to improve the performance. Finally, this thesis gives some experimental results comparing a few of the algorithms discussed in this thesis, which demonstrates the effectiveness of the proposed Adaboost.ML algorithm.In other hand, the study of multi-label classification is more and more. Multi-label learning mainly solves the problem of a sample with multi-label, and it is suitable for all kinds of classification task. In traditional multi-label learning methods, classifiers are usually required to utilize a large amount of fully labeled training data in order to obtain good performances for multi-label classifications. But in many real tasks, it is much easier to obtain partially labeled training data, and it costs less efforts than obtaining a large amount of fully labeled training data. This thesis presents a multi-label classification method for weak labeling that is RPCME algorithm, which deals with data by using the way of pairs constraints projection based on similarity. The method can make better use of characteristics of the sample with weak label and improve the classification performance.This thesis has studied Adaboost algorithm based on stable classier and multi-label ensemble classification for weak label, and presented two algorithms named Adaboost.ML and RPCME, which provides meaningful thoughts for future researches on such kind of problems.