József Valyon, Gábor Horváth
Least Squares Support Vector Machines for Data Mining
Due to the extensive use of databases and data warehouses; it is very easy to collect large quantities of data from any aspect of life. The analysis of these data may lead to many useful or interesting conclusions. Data mining is a collection of methods and algorithms that can effectively be used to discover implicit, previously unknown, hidden trends, relationships, patterns in masses of data. Classical data mining uses many methods from different fields, like the foundations of linear algebra, graph theory, database theory, machine learning and artificial intelligence. In this paper, we address the problem of black-box modeling, where the model is built based on the analysis of input-output data. These problems usually cannot be addressed analytically, so they require the use of soft computing methods. We focus on the use of a special type of Neural Network (NN), the least squares version of Support Vector Machines (SVMs), the LS-SVM. We also present an extension of this method, the LS2-SVM, which enables us to deal with large quantities of data. Using this method we present solutions for function approximation, time series analysis, and time series prediction.