Studies On Semi-supervised Learning And Its Applications

The paper is about semi-supervised learning and its applications in data mining technology. The knowledge of machine learning, data mining, their histories, and processes are introduced. Also, we use semi-supervised machine learning algorithms. At last, we give out data mining applications, for useful information.Machine learning is used in data mining technology. Semi-supervised learning is exploiting both labeled and unlabeled data. Supervised learning assumes the class labels are known and its algorithms require sufficient training data so that the obtained model can generalize well. However, data in many domains are unlabeled; obtaining unlabeled data is rather easier. Unsupervised learning makes no assumptions about the category structure, which makes it more difficult for machine learning process. Semi-supervised learning can be beneficial to traditional supervised learning and unsupervised learning. The paper describes utilizing both labeled and unlabeled samples in building classifiers and clustering methods. The major innovations and research works are as followed:(1) The paper studies the technical bases and related works on semi-supervised learning. Compared with traditional supervised learning and unsupervised learning, semi-supervised learning is in a rather new field. Also, the paper has studied the related works on data mining.(2) The paper gives out a semi-supervised learning algorithm for Bayesian classification. The algorithm is based on Bayesian decision machinery, with estimation of the structure from probability density function, for two-class problems.(3) The paper gives out a semi-supervised learning algorithm based on FCM. The algorithm uses the idea of unsupervised clustering, indirected measurement of separating clusters, fuzzy method, and feature selection, for multiple-class problems.(4) The paper gives out semi-supervised learning algorithm as technical solution for data mining model in credit card application. It uses semi-supervised learning algorithm based on FCM, and for feature selection. Concerned with the characteristics of credit card approval model, cost function is applied, for recognizing different users. Also, the paper gives out Magic telescope data analysis application, for separating high-energy ray signal with background.