Blood Cell Classification Based on Feature Importance with the Naïve Bayes Algorithm on the BCCD Dataset

Abstract

Classification of blood cells is of immense importance as a preliminary stage in the diagnosis of several hematological disorders such as leukemia, anemia, and infections. Microscopes and trained personnel often constrain the manual approach due to limited resources, time, and the ability to produce consistent results. In this work, BCCD (Blood Cell Classification Dataset) has been used to solve this problem. This dataset contains 364 images to classify three types of blood cells, which are red blood cells (RBC), white blood cells (WBC), and platelets using the Naïve Bayes algorithm. Feature extraction includes color (RGB), intensity (grayscale), texture (GLCM), shape, and edge density. The data set is split into 80% for training and 20% for testing. Results of the study showed that the Naive Bayes model obtained an accuracy of 98.64% on the test data. The precision of platelets, WBC, and RBC was 85%, 97%, and 100%, respectively. The recalls for the three classes were 98%, 99%, and 99%, respectively. The study demonstrated that the machine learning methods, particularly the Naive Bayes algorithm, constitute a very viable alternative to blood cell classification because of high computation speed, robust performance, and the capability of dealing with high-dimensional data under the simple assumption of feature independence.