In this Research paper, Yang et al. proposed a heuristic for coping with the random masking of the values in the data matrix. To address this issue and to further accelerate the biclustering process, the authors presented a new model of bicluster to incorporate null values seamlessly. This bicluster model was proposed to capture coherence among a subset of the attributes and coherent behavior rather than points/objects that are physically close to each other. Due to the NP-hard nature of this problem, Author devise a new move-based algorithm (FLOC) that can efficiently approximate the clusters with high accuracy. This algorithm can be easily modified to discover clusters with different constraints, e.g. overlapping or no overlapping clusters simultaneously based on probabilistic moves. The algorithm begins with the creation of k initial biclusters with rows and columns added to them according to a given probability. After that, these biclusters are iteratively improved by the addition or removal of one row or column at a time, determining the action (add/removal of column i.e. Action(x, c), where x is column & c is cluster) that better improves the average of the MSR (Mean Squared residue: The residue is a measurement of the degradation to the coherence of the cluster that an entry brings) values of the k biclusters. The best clustering obtained during each iteration will serve as the initial clustering for the next iteration. The algorithm terminates when the current iteration fails toImproves the overall clustering quality. The smaller the Residue, the stronger the coherence. Our objective is then to find clusters that minimize the residue value. The order of Action, for the sake of Cluster quality optimization, is taken into consideration as well. One is Static ordering & other is dynamic. The drawback of static ordering is that If a large number of actions with negative gain proceed a small number of actions with positive Gain(reduction of c’s residue incurred by performing Action(x,c)) at the end of the performance list, then the set of actions with positive gain may never be given a full play. Bicluster volumes (The number of specified entries in the corresponding submatrix is referred to as the volume of the cluster) are also taken into account within the possible actions, where bigger biclusters are preferred, and the variance is used to reject constant biclusters (When the variance of the embedded clusters increases, the volume of the embedded clusters becomes more different. On the other hand, the volume becomes more homogeneous with smaller variance. All clusters have the same volume if the variance is0). A parameter ? is introduced to control the size of a ? -cluster. The average number of rows and columns in a cluster is ?×N ?×M respectively. By experimenting with different value of ?, if the optimal cluster size is very different from the initial (seed) cluster size, then it would take more iterations to reach the optimal cluster and the response time could be prolonged. In order to accelerate the response time, it is beneficial to make the initial seed as close to the optimal cluster size as possible. In order to properly accommodate the object/attribute biases within a ? – cluster, author introduced a concept base to represent the bias of an object or attribute within a ?-cluster. For a given ? -cluster (I, J), the base of an object Oi is defined as the average value of Oi for all specified attributes in J , is the set of specified attributes in J for object Oi