Factor analysisin data mining

In this study, we propose a method of factor analysis for a huge database so that not only the independence among the factors can be ensured and the levels of their importance can be measured, but also new factors can be discovered. To measure the independence of the factors, statistical relation analysis and the concept of fuzzy set theory are employed. A fuzzy set of ‘the factors are almost dependent’ is used to measure the degree of dependence between factors, and then through a hierarchical clustering procedure, the dependent factors are detected and removed. To measure the weights of importance for the independent factors, a supervised feedforward neural network is developed. In addition, we design a hierarchical structure to facilitate the extraction of new factors when the information of the system is not complete. The applicability of the proposed model is evaluated by a case of customersʹ contribution analysis of a telecom company with 8% error rate.