Programming measurements was utilized for foreseeing issue in modules of programming ventures. Convenient forecast of flaws enhances programming quality and subsequently its dependability. In this paper, a framework towards subspace grouping of large data set was proposed at class level to minimize the error. We composed an iterative calculation for grouping of high dimensional datasets for improvement of a target work. At that point the bunched data sets were examined utilizing Step-Wise Linear Regression to investigate the relationship among a structure variable and the autonomous factors in order to anticipate of damaged and non-faulty classes. To evaluate the supportiveness of the model, we drove a practical learning on the Attitude Survey Data. The proposed strategy specifically managed blunder variables and consequently gave precise fault prediction least standard error (0.003) when contrasted with the current technique (4.687). Root mean square error which measures the distinction between the assessed error and the real error was (0.8) in the proposed technique. The results demonstrated that the forecast models based on subspace clustering were essentially predominant to the current techniques. © 2018 Sumangala Patil et. al.