Supervised Machine Learning Approaches for Breast Cancer Classification and a high performance Recurrent Neural Network

Breast Cancer is among the most widespread forms of cancer that affects hundreds of thousands of women across the world each year. Early detection of this cancer can really help, in that it cannot only make the treatment easier, but also significantly improve survival chances. It is thus imperative that the machine learning algorithm being deployed for real world use be reliable and accurate. This paper not only compared the performance of several classifiers on the given dataset on breast cancer from UCI with data from close to 600 cases, but also used the Recurrent Neural Network (RNN) for classification, which to the best of the authors' knowledge has never been used before for breast cancer classification. Also, the performances of these classifiers are compared with the same already implemented before in this domain and discuss about the improvements made in the proposed work. The proposed experimental results showed that, RNN has yielded best performance with the highest accuracy of 98.49%, followed by the linear kernel SVM classifier with an accuracy of 98.24% and XGBoost at 97.36%. This research work has trained and tested a total of 9 classifiers and analyzed their performance by considering accuracy as the key metric. Although, to the best of the authors' knowledge, the Recurrent Neural Network has not been implemented in this domain before and is thus novel. Since our domain is mission critical, the proposed research work highly emphasizes on accuracy, while ensuring that the every classifier is trained and tested accurately.