Clustering Cells Shape Descriptors Using K-Means vs. Genetic Algorithm

Abstract

This paper interested in clustering red blood cells, these cells are in form of digital images of blood films, a comparison made between Genetic Algorithm (GA) and K-Means behavior/performance in clustering. The data set consists of shape descriptors of the cells shapes, the original number of samples are 100 samples. Each sample provided us with at least 10 cells (shape) with total number of 409 shapes (cells). The Genetic Algorithm shows better performance than K-Means in clustering these cells into two clusters (Normal and Abnormal) with success rate 99.48% where K-Means gave 83.16%. While K-Means shows a better performance in clustering the cells into four clusters (Burr, sickle, teardrop and normal cells) than GA where K-Means gave 86.74% and Genetic algorithm (GA) gave 83.2 %.