Visualizing breast cancer data with t-SNE

One in eight women will get breast cancer in her lifetime and in 2008 it has caused 458.503 deaths among the world [15]. Despite that technology has made considerable improvements in the last decades, there is still room for more advances. A technique that possibly can contribute to this field is t-SNE [24]. The aim of this thesis is to investigate whether t-SNE is able to present the breast cancer data in an interpretable way and possibly improves the classification performances. We employ two approaches to explore the applicability of t-SNE. In the first approach we compare the visualizations and in the second approach the classification performances are compared. We found that classification on the original data performed significantly better than on t-SNE data. This suggests that t-SNE is not applicable to the breast cancer data set.