Seim-supervised Classification Based On Extreme Learning Machine

Traditional supervised learning and unsupervised learning are two frequently-usedlearning methods in the field of machine learning．In supervised learning，a largenumber of labeled data are taken as prior knowledge to construct a model which isused to predict the unlabeled data．Unsupervised learning is always absence of anyprior knowledge to analyze the data and complete clustering. With the rapiddevelopment of information technology, collection of a large amount of unlabeled databecome quite easy, however, because it requires some human resources and materialresources, gaining labeled data is relatively difficult. If only a few labeled data are used,then the learning model training by supervised learning does not have very goodgeneralization ability, and at the same time, a lot of unlabeled data are wasted; If onlythe unlabeled data are used, then the unsupervised learning will ignore the value oflabeled data.Therefore, semi-supervised learning both making use of a small amount oflabeled data and a large number of unlabeled data to improve the performance oflearning machine has become one of the hotspots in the research of machine learning.When the training sample(labeled data) is not enough, semi-supervised classificationcan use the information of a large number of unlabeled data to improve theperformance of the classifier.A novel machine learning algorithm, namely, Extreme learning machine(ELM) isput forward in recent years. Different with the traditional neural network model,Extreme learning machine requires activation function infinite order differentiable,with a random assignment to the input weights and partial values of the network, thenit trains hidden layer of the network by the least square method, finally it calculates theoutput weights of the network using the generalized inverse, the computing speed isfast. With the advantages of ELM, now it is well applied in the Function approximation,time series prediction, pattern recognition and classification. However, when extremelearning machine applied in classification problems, as the number of labeled datadecreases, the accuracy of the classification model training by labeled data will decrease.When getting labeled data is very difficult, and at the same time there are manyunlabeled data, using the advantages of machine learning algorithm andsemi-supervised learning, this paper proposed the seim-supervised classificationalgorithm based on extreme learning machine, the main work is as follows:(1)When the data set has fewer labeled data, the lower classification accuracy ofthe classification model training by the Extreme Learning Machine, this chapter usingFuzzy C mean cluster algorithm and extreme learning machine algorithm makes use ofa large number of unlabeled data and fewer labeled data to achieve semi-supervisedclassification. When datasets have fewer labeled data, the classification accuracy ofthis method is significantly better than the traditional unsupervised clusteringalgorithm FCM and supervised Extreme Learning Machine classification algorithm.(2)Fuzzy C-Means cluster algorithm’s shortcoming is the sensibility to initialvalue，it usually leads to local minimum. The global Fuzzy C-Means clusteringalgorithm can give us more satisfactory results by escaping from the sensibility toinitial value and improv the accuracy of clustering. So this chapter we use the globalFuzzy C-Means clustering algorithm and Extreme Learning Machine to achievesemi-supervised classification.(3)To improve the generalization performance, differential evolution algorithm，which has the features of global convergence and easy computation, is introduced inthe parameter optimization of extreme learning machine.The performance ofsemi-supervised classification with evolution ELM promoted further.