Dimension reduction is the process of keeping only those dimensions in a dataset which are important from the point of view of problem at hand and discarding of the others. This helps to design easily computable algorithms and to increase the performance of classifiers. It has gained importance as a preprocessing step in knowledge discovery and data mining especially in the fields of pattern matching, machine learning, bioinformatics and genetics which involve datasets having large number of dimensions. There are two basic strategies used for reduction of the dimensions; feature selection and feature extraction. Feature selection techniques focus on selecting some of the important features from all the features while feature extraction techniques are based on generating new features by making use of entire information present in the original dataset. Recently some work has also focused upon combining both the strategies to club their advantages. This paper studies the basic strategies for dimension reduction, the different techniques proposed in literature for reducing the dimensions and also about the measures used by them. Finally it also discusses about the combined approach and concluding remarks are given.

Dimension reduction Feature Selection Feature Extraction.