Breast Cancer is a non-communicable disease seen primarily in women population. As per the statistics published by the World Health Organization, it is presently ranked, globally, as number one in incidence. It can principally affect women of any age, and can be diagnosed in any of the five stages of the disease, but chances of cancer survival become more difficult when diagnosis is made in advanced stages of the disease. Mortality rate of cancer is seen to be high in developing countries than in developed countries. Owing to this fact breast cancer prediction, diagnostic and therapeutic facilities need to be urgently improved in this extent. Henceforth, development of clinical decision support systems for early and precise detection of the disease gains significance and is the need of the hour. The study aims in building a model for precise classification of breast tumors with minimum misclassification of labels. In this paper the potential of extra tree classifiers for breast cancer classification into malignant or benign tumors is examined. A model for breast cancer classification is proposed using extremely randomized tree classifier. Hyperparameter optimization is applied. Identification of important features aids in model performance. Features relevant to disease detection are identified and ranked by importance using 3 techniques- impurity based, permutation based and Shap values. The most important four features identified are Size Uniformity, Shape Uniformity, Bare Nucleoli and Normal Nucleoli. Performance of the optimized model is analyzed using training-testing partitions and k fold stratified cross validation with k as 5 and it was observed that they produced an accuracy of 99.27% on the test set and 97.3 % on the cross validated model respectively. The study reveals the suitability of the extra tree classifier for breast cancer classification. The model is compared with other state of art models and it was seen to be superior in performance. Furthermore, extremely randomized tree classifiers are perceived to be suitable in developing models for breast cancer classification with minimal misclassification of instances.