Disease-Level Detection based on Blood Cells and Machine Learning Models

Abstract

Blood cells transmit critical information that can be used to determine an individual's current health status. As a result, this facilitated the identification of diseases that are linked to blood cells. It is imperative to mitigate the infection risks that individuals face on a daily basis by utilizing blood cell analysis to detect numerous diseases in an immediate and precise manner. The evaluation, categorization, and enumeration of biological entities necessitate a significant amount of time, and the reliance on subjective human judgment may introduce inaccuracies. Machine learning approaches are effective for simple classification challenges. In this study introduce traditional machine learning models for detection of disease based on blood cells dataset. Mainly, three ML models are applied in the detection process which are k-Nearest Neighbors (KNN), Logistic Regression (LR), and Xgboost. Furthermore, 19 features are used in feature engineering process and as input in the training and testing process of proposed study. Selected blood cells dataset is used to detect six main cases which are Anemia, Artefact, Infection, Leukemia, Normal, and Sickle_Cell. Different hyper-parameters settings are applied during experiments. Results of experiments are carried out performance evaluation for proposed study based on different ML assessment metrics and comparison with other ML models. The most promising results based on selected features and three ML models is related with Artefact class. However, results show that Xgboost classifier is 96% accurate in detecting diseases from blood cell dataset and can be valuable practically in medical sector.