A Novel Computational Method to Find Tumour Nuclei Traits Predictive of Breast Tumour Malignancy Using Machine Learning

Breast cancer is one of the most common cancers in the world, where early detection and diagnosis greatly improves survival. In this paper, we propose a decision-tree based model operated on the Wisconsin Breast Cancer Dataset for the breast cancer tumour classification problem. We trained three machine learning classification models, namely a mean nuclei measurements model to represent nuclei size, a standard error (SE) nuclei measurements model to represent nuclei irregularity, and a model on all the data, since cancer cells are known to have large and irregular nuclei. In this paper, we will show that the mean nuclei measurements model provides the best accuracy and sensitivity in classifying malignancy. For specificity, the mean nuclei measurements model and full dataset model perform better than the SE nuclei measurements model.