Imbalanced Fuzzy Weighted Extreme Learning Machine And Its Bagging Ensemble Methods

With the rapid development of information science and technology, the generation and storage of data become extremely simple and fast, then how to extract useful information and knowledge from such a large amount of data is an urgent problem to be solved. The purpose of data mining technology is to search for hidden information and knowledge from a large number of data through the powerful algorithms, effectively improving the utilization of a large number of idle data.Classification task which determines the sample belonging to which predefined target class, is one of the core technologies of data mining. At present, the development of classification technology has become more mature, but most of the traditional classification algorithms are based on the balanced data. However, raw data with imbalance class distribution can be found almost everywhere. Most of the classifiers can produce an undesirable model that is biased toward the majority class and has a low performance on the minority class. Extreme learning machine has the advantages of high classification accuracy and light computational requirements. When it is used to solve the problem of unbalanced classification, the classification results affected by the imbalanced distribution of data can be poor. Focusing on these problems, the main work of this thesis is as follows:(1)The study of class imbalance fuzzy weighted extreme learning machine:by combining the distribution characteristics of imbalanced data sets and the structural mechanism of extreme learning machine, the negative effects caused by the unbalanced class distribution is theoretically proved. We also discuss three factors, i.e., unbalanced ratio, sample size, noise and how these factors influence the performance of the extreme learning machine. Furthermore, the prior distribution of the training data is fully extracted, and the algorithm of the fuzzy weighted extreme learning machine is proposed. Experimental results show that compared with the weighted extreme learning machine and some other extreme learning machine algorithms, the fuzzy weighted extreme learning machine can obtain better classification performance. In contrast with the FSVM-CIL, it can obtain the similar classification performance, but the time cost is much smaller.(2)The study of FWELM-based bagging ensemble methods:A bagging ensemble learning algorithm of class imbalance fuzzy weighted extreme learning machine(FWELM) is proposed to overcome the drawbacks of the FWELM:low stability and low generalization ability. The proposed algorithm generates lots of FWELM classifiers to embed into the Bagging integrated learning model, which by the strategy of randomly assigning input weights and biases of the hidden layer. Experimental results on 15 skewed data sets indicate that the Bag-FWELM algorithm is more accurate, robust and efficient.