A Stochastic Neighbor Embedding Approach for Cancer Prediction

Abstract

Nowadays, cancer is the most prevalent cause of death worldwide. Breast cancer is the 2nd most common cancer registered globally and the leading death factor among women. So, if we predict cancer in the primitive stage, we can avoid death. Therefore, there is an imperative need of a tool that can quickly detect whether the breast cancer tumor is a malignant or a benign one. In this paper, two-dimensionality reduction methods, namely Principal Component Analysis (PCA) and t-Distributed Stochastic Neighbor Embedding(t-SNE) are used. t-SNE visualizes high-dimensional data by reducing it in a two or three-dimensional map. It optimizes the cost function and produces much better scatter plots by decreasing the tendency to the crowd points in the center of the map. For classification, two machine learning algorithms, namely Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost), are employed. The experimental result reveals that t-SNE performed better as compared to PCA.