The breast cancer in the women is most commonly diagnosed type of cancer. The mortality rate can be reduced if proper and early breast cancer treatment can be made. Masses and microcalcification contain very important diagnostic information in breast cancer. There is great variation in masses and micro-calcifications so, radiologists face difficulties in proper diagnosis of the breast cancer. Researchers in the past developed efficient systems based on computer aided diagnostic (CAD) systems. Moreover, relevant feature extraction plays a vital role in proper diagnostic and prognostic. Based on the diverse nature and variations in the breast cancer mammograms, we propose hybrid feature extraction approach including morphological, entropybased features, elliptic Fourier descriptors (EFDs), texture and scale invariant feature transform (SIFT). For improving the detection accuracy based on the extracted features, we applied machine learning classifiers including Support vector machine (SVM) alongwith its kernels such as Gaussian, radial base function (RBF), polynomial; Na?ve Bayes and Decision tree (DT). The validation was measured using 10-fold cross validation (CV) system. For a performance evaluation, we computed different metrics including specificity, sensitivity, false positive rate (FPR), negative predictive value (NPV), positive predictive value (PPV), and area under the receiver operating curve (AUC). Both combination and single feature are used as an input for classifiers. The results reveal that both single and combination of features provides higher detection results. Thus, the new feature extracting approach is more robust in early detection of breast cancer.