Classification techniques function as a main component in digital mammography for breast cancer treatment. While many classification techniques currently exist, recent developments in the derivatives of Support Vector Machines (SVM) with feature selection have shown to yield superior classification accuracy rates in comparison with other competing techniques. In this paper, we propose a new classification technique that is derived from SVM in which margin is maximised and redundancy is minimised during the feature selection process. We have conducted experiments on the largest publicly available data set of mammograms. The empirical results indicate that our proposed classification technique performs superior to other previously proposed SVM-based techniques.