CLASSIFICATION OF SICKLE CELL ANAEMIA SEVERITY USING HAEMATOLOGICAL PARAMETERS: A MACHINE LEARNING APPROACH

ABSTRACT

Sickle Cell Anaemia (SCA) is a genetic blood disorder that caused abnormal haemoglobin and severe health complications. In Nigeria, nearly 150,000 infants, 2% of newborns, were diagnosed annually, highlighting its public health impact. SCA severity varied and was linked to complications like vaso-occlusive and other critical case. Traditional severity assessments were subjective, but machine learning (ML) provided an objective alternative. This study developed an ML-based severity classification model using haematological parameters from 364 Complete Blood Count (CBC) patient records. Features included age, sex, RBC, PCV, MCV, MCH, MCHC, RDW, TLC, PLT/mm³, and HGB. ML models including Logistic Regression, Support Vector Machine, Random Forest, and XGBoost were tested with Chi-Square, ANOVA, and all features. Logistic Regression achieved the highest accuracy at 96%, Random Forest reached 95%, XGBoost ranged from 89% to 92%, and Support Vector Machine remained at 90%. ANOVA proved the best feature selection method. Future work should aim at using ensemble learning for enhanced SCA severity prediction.

Keywords: Feature Selection, Machine Learning Models, Model Performance, Severity Classification, Sickle Cell Anaemia.