<p><span>In recent times Big Data Analysis are imminent as essential area in the field of Computer Science. Taking out of significant information from Big Data by separating the data in to distinct group is crucial task and it is beyond the scope of commonly used personal machine. It is necessary to adopt the distributed environment similar to map reduce paradigm and migrate the data mining algorithm using it. In Data Mining the partition based K Means Clustering is one of the broadly used algorithms for grouping data according to the degree of similarities between data. It requires the number of K and initial centroid of cluster as input. By surveying the parameters preferred by algorithm or opted by user influence the functionality of Algorithm. It is the necessity to migrate the K means Clustering on MapReduce and predicts the value of k using machine learning approach. For selecting the initial cluster the efficient method is to be devised and united with it. This paper is comprised the survey of several methods for predicting the value of K in K means Clustering and also contains the survey of different methodologies to find out initial center of the cluster. Along with initial value of k and initial centroid selection the objective of proposed work is to compact with analysis of categorical data.</span></p>