Breast Cancer Detection Using K-Nearest Neighbour Algorithm
Shagun Chawla, Rajat Kumar, Ekansh Aggarwal, Sarthak Swain · SSRN Electronic Journal · 2018
Breast cancer is one of the common occurring cancer in women across the globe, affecting about significant percentage of women at some point in their life. Even with the development of new technologies in the field of medicine and research, the accurate diagnosis of this fatal disease outcome is one of the most important tasks needed to be done till date. Our objective is to develop a sophisticated and automated diagnostic system that yields accurate and reproducible results for predicting whether a breast cancer tumour is benign (non-cancerous) or malignant (cancerous). We have implemented KNearest Neighbour Algorithm using various normalization techniques and distance functions at different values of K. A comparative study using various normalization techniques, i.e., Min-Max normalization, Z-Score normalization and Decimal Scaling normalization, and different distance metrics, i.e., Manhattan distance, Euclidean distance, Chebyshev distance and Cosine distance has been done. The accuracy of each variation is tested and the maximum accurate prediction is considered for the result. Highest accuracy of 98.24% is achieved, with KNN implementation using Manhattan distance metric, at K=14, along with Decimal scale normalization.