Fast Feature subset selection algorithm based on clustering for high dimensional data

A Feature selection algorithm employ for removing irrelevant, redundant information from the data. Amongst feature subset selection algorithm filter methods are used because of its generality and are usually good choice when numbers of features are large. In cluster analysis, graph-theoretic clustering methods to features are used. In particular, the minimum spanning tree (MST)based clustering algorithms are adopted. A Fast clustering bAsed feature Selection algoriThm (FAST) is based on MST method. In the FAST algorithm, features are divided into clusters by using graph-theoretic clustering methods and then, the most representative feature that is strongly related to target classes is selected. Features in different clusters are relatively independent. A feature subset selection algorithm (FAST) is used to test high dimensional available image, microarray, and text data sets. Traditionally, feature subset selection research has focused on searching for relevant features. The clustering-based strategy of FAST having a high probability of producing a subset of useful and independent features. Keywords— Cluster analysis, Graph-theoretic clustering, Minimum spanning tree, Feature selection, feature subset selection algorithm (FAST), High dimensional data, Filter method. INTRODUCTION Data mining is a process of analyzing data and summarizes it into useful information. In order to achieve successful data mining, feature selection is an essential component. In machine learning feature selection is also known as variable selection or attributes selection. The main idea of feature selection is to choose a subset of features by eliminating irrelevant or no predictive information. It is a process of selecting a subset of original features according to specific criteria. Feature selection is an important and frequently used technique in data mining for dimension reduction. It employ for removing irrelevant, redundant information from the data to speeding up a data mining algorithm, improving learning accuracy, and leading to better model comprehensibility. Supervised, unsupervised and semi-supervised feature selection algorithms are developed as result of process of feature selection algorithm. A supervised feature selection algorithm determines features' relevance by evaluating their correlation with the class or their utility for achieving accurate prediction, and without labels, an unsupervised feature selection algorithm may exploit data variance or data distribution in its evaluation of features' relevance and a semi-supervised feature selection algorithm uses a small amount of labelled data as additional information to improve unsupervised feature selection [2]. Feature subset selection methods can be divided into four major categories: Embedded, Wrapper, Filter, and Hybrid. The embedded methods has a feature selections as a part of the training process and are usually specific to given learning algorithms, and thus possibly more efficient than the other three categories. Machine learning algorithms like decision trees or artificial neural networks are examples of embedded approaches. Wrapper methods assess subsets of variables according to their relevance to a given predictor. The method conducts a search for a good subset using the learning algorithm itself as part of the evaluation function. Filter methods are pre-processing methods. They attempt to assess the useful features from the data, ignoring the effects of the selected feature subset on the performance of the learning algorithm. Examples are methods that select variables by ranking them through compression techniques or by computing correlation with the output. The hybrid methods are a combination of filter and wrapper methods by using a filter method to reduce search space that will be considered by the subsequent wrapper. The important part of hybrid method is combination of filter and wrapper methods to achieve the best possible performance with a particular learning algorithm with similar time complexity of the filter methods [1]. In cluster analysis, graph theoretic approach is used in many applications. In general graph-theoretic clustering a complete graph is formed by connecting each instance with all its neighbours. Zahn's clustering Algorithm 1. Construct the MST for the set of n patterns given. 2. Identify inconsistent edges in MST. 3. Remove the inconsistent edges to form connected components and call them clusters. In the FAST algorithm, features are divided into clusters by using graph-theoretic clustering methods and then, the most representative feature that is strongly related to target classes is selected. Features in different clusters are relatively independent. A feature subset selection algorithm (FAST) is used to test high dimensional available image, microarray, and text data sets. Traditionally, feature subset selection research has focused on searching for relevant features. The clustering-based strategy of FAST International Journal of Engineering Research and General Science Volume 2, Issue 6, October-November, 2014 ISSN 2091-2730 768 www.ijergs.org having a high probability of producing a subset of useful and independent features.