Adaptive dimension reduction using discriminant analysis and K-means clustering