Predicting the Severity of Breast Masses with Ensemble of Bayesian Classifiers

Abstract

Problem statement: This study evaluated two different Bayesian classifiers; tree augmented Naive Bayes and Markov blanket estimation networks in order to build an ensemble model for prediction the severity of breast masses. The objective of the proposed algorithm was to help physicians in their decisions to perform a breast biopsy on a suspicious lesion seen in a mammogram image or to perform a short term follow-up examination instead. While, mammography is the most effective and available tool for breast cancer screening, mammograms do not detect all breast cancers. Also, a small portion of mammograms show that a cancer could probably be present when it is not (called a false-positive result). Approach: Apply ensemble of Bayesian classifiers to predict the severity of breast masses. Bayesian classifiers had been selected as they were able to produce probability estimates rather than predictions. These estimated allow predictions to be ranked and their expected costs to be minimized. The proposed ensemble used the confidence scores where the highest confidence wins to combine the predictions of individual classifiers. Results: The prediction accuracies of Bayesian ensemble was benchmarked against the well-known multilayer perceptron neural network and the ensemble had achieved a remarkable performance with 91.83% accuracy on training subset and 90.63% of test one and outperformed the neural network model. Conclusion: Experimental results showed that the Bayesian classifiers are competitive techniques in the problem of prediction the severity of breast masses.