Regression analysis for high-dimension and low-sample size (HDLSS) data has problem that we cannot obtain result of predicted values of dependent variables. This paper presents a new methodology used to solve this problem. Data is classified into several subclasses of data sets. Following that, we use a technique which can obtain the predicted values of dependent variables, in the same linear subspace, over the several subclasses of the data sets. For capturing the same linear subspace, we utilize a common scale obtained as fuzzy clusters in a fuzzy clustering result over the subclasses. Numerical examples, using the HDLSS data, show better performance when compared with the ordinary regression analysis.