Factor Analysis in Data Mining

The rapid growth and advances of information technology enable data to be accumulated faster and in much larger quantities (i.e., data warehousing). Faced with vast new information resources, scientists, engineers, and business people need efficient analytical techniques to extract useful information and effectively uncover new, valuable knowledge patterns. Data preparation is the beginning activity of exploring for potentially useful information. However, there may be redundant dimensions (i.e., variables) in the data, even after the data are well prepared. In this case, the performance of data-mining methods will be affected negatively by this redundancy. Factor Analysis (FA) is known to be a commonly used method, among others, to reduce data dimensions to a small number of substantial characteristics. FA is a statistical technique used to find an underlying structure in a set of measured variables. FA proceeds with finding new independent variables (factors) that describe the patterns of relationships among original dependent variables. With FA, a data miner can determine whether or not some variables should be grouped as a distinguishing factor, based on how these variables are related. Thus, the number of factors will be smaller than the number of original variables in the data, enhancing the performance of the data-mining task. In addition, the factors may be able to reveal underlying attributes that cannot be observed or interpreted explicitly so that, in effect, a reconstructed version of the data is created and used to make hypothesized conclusions. In general, FA is used with many data-mining methods (e.g., neural network, clustering). BACKGROUND