LIFE EXPECTANCY PREDICTION FOR SICKLE CELL DISEASE PATIENTS USING EXPLAINABLE MACHINE LEARNING

Predicting mortality risk among patients with Sickle Cell Disease (SCD) remains challenging because survival is influenced by interacting demographic, clinical, laboratory, treatment, and healthcare-related factors. This study developed an explainable machine learning framework for survival risk prediction among patients with SCD in Nigeria. The study used a dataset of 1,000 patients with 51 demographic, clinical, laboratory, treatment, and survival-related variables. Logistic Regression, Random Forest, and Extreme Gradient Boosting (XGBoost) models were trained following preprocessing and feature engineering. The dataset was divided using an 80:20 stratified train-test split to preserve the minority outcome class, while class weighting was incorporated into the classification models to address class imbalance. Model performance was evaluated using accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC), with F1-score emphasized because of the substantial class imbalance. Random Forest and XGBoost achieved the highest test accuracy (0.97), compared with 0.81 for Logistic Regression. For the minority class representing high risk of early lifespan, Random Forest and XGBoost achieved precision of 1.00, recall of 0.143, and F1-score of 0.250, whereas Logistic Regression achieved recall of 0.286 and F1-score of 0.095. Test AUC was highest for Logistic Regression (0.731), followed by Random Forest (0.679) and XGBoost (0.663). Random Forest was selected as the best-performing model based on F1-score and cross-validation results. SHAP analysis of the XGBoost model identified demographic characteristics, laboratory measures, disease-related complications, and follow-up factors as important contributors to survival predictions. The findings indicate that ensemble models can provide useful risk stratification for SCD, but their low minority-class recall highlights the need for improved imbalance-handling strategies before clinical deployment. The integration of SHAP provides clinically interpretable information that can support individualized risk assessment and monitoring in Nigerian SCD care.