Frequency Count Based Filter for Dimensionality Reduction

Selecting relevant features from a dataset has been considered to be one of the major components of data mining techniques. Data mining techniques become computationally expensive when used with irrelevant features. Dimensionality reduction/feature selection algorithms are used basically to reduce the dimension of a dataset without reducing the information content of the domain. There are basically two categories of feature selection methods. Supervised, where each instance is associated with a class label, and in unsupervised, instances are not related to any class label. Unsupervised feature selection is used as a pre-processing of other machine learning techniques such as clustering, classification, association rule mining to reduce the dimensionality of the domain space without much loss of information content. This paper presents an unsupervised dimensionality reduction technique from continuous valued dataset, based on frequency count.