Human Blood Type Classification Using Feature Selection and Optimized Machine Learning Model

Blood type classification, a fundamental aspect of medical diagnosis is traditionally and manually administered by registered skilled medical technologists. In this study, the proponent developed an intelligent model for human blood typing using image processing, feature extraction, and advanced machine learning methodologies. The dataset underwent a process from image acquisition, image processing, segmentation, binary conversion, and feature extraction. Univariate feature selection and the chi2 algorithm were employed to refine the original thirty-nine features down to the top twenty features. Seven machine learning algorithms were explored using default parameters. The models were then optimized using GridSearch to further enhance their performances. In LR, Blood types O+, A-, and B+ displayed the lowest F1 scores, yielding a macro-average of 89%. Meanwhile, KNN showed its lowest F1 score of 77% for Blood type B+, contributing to a macro-average of 91%. Similarly, SVM manifested the lowest F1 score for Blood type B+, resulting in a macro-average of 90%. Increasing the size of the training set size and validation set size should be taken into consideration in order to further enhance model performance. This study contributes insights into developing and optimizing intelligent models for blood typing and potential avenues for future research and model enhancements by integrating cutting-edge technologies.