Abstarct: Clustering is one of the important data mining issues for discovering hidden patterns in data. If the number of variables are too much and we encounter with high dimensional data , the problem of the dimension reduction is paralled to clustering or in the same direction of it. One of the most commonly methods of dimension reduction, which is used in both monitoring and uncontrolled learning topics, is the principal component method that has its own merits and disadvantages. In this thesis we trying by introducing optimal discriminant clustering (ODC) method, that is used to reduce the dimension, describe the clustering problem, which is an uncontrolled learning problem, as a problem with ridge regression, so that, like the thought of the principal components, one can extrapolate another kind of linear combination of initial variables to construct new variables, then use one of the algorithms k-means clustering for new converted observations. Since the ridge regression problem plays prominent a role in the Tuning parameter, In this case, an tuning parameter will play a significant role in clustering performance. Also, the existence of some nonessential variables in the model leads to negativ and weak performing of clustering method, so by adding the group’s fine fractional function, we will eliminate this weakness and introduce a new clustering, that is the revised version of the optimal discriminant clustering. The results of the simulation indicate the effectiveness of this method in dealing with the high dimensions of the variables as well as its superiority to the principal components analysis method.