Kernel-based fuzzy-rough nearest-neighbour classification for mammographic risk analysis

Mammographic risk analysis is an important task for assessing the likelihood of a woman developing breast cancer. It has attracted much attention in recent years as it can be used as an early risk indicator when screening patients. In this paper, a kernel-based fuzzy-rough nearest-neighbour approach to classification is employed to address the issue of the assessment of mammographic risk. Four different breast tissue density assessment metrics are employed to support this study, and the performance of the proposed approach is compared with alternative nearest-neighbour-based classifiers and other popular learning classification techniques. Systematic experimental results show that the work employed here generally improves the classification performance over the others, measured using criteria such as classification accuracy rate, root mean squared error and the kappa statistics. This demonstrates the potential of kernel-based fuzzy-rough nearest-neighbour classification as a robust and reliable tool for mammographic risk analysis.