Using explainable machine learning to identify predictors of Kangaroo mother care implementation in Sierra Leone's healthcare system

• XGBoost accurately predicts the adoption of Kangaroo Mother Care (KMC) in Sierra Leone. • Explainable AI identifies key clinical and socio-economic drivers of KMC uptake. • SHAP and LIME provide transparent, actionable insights for healthcare decisions. • Model supports targeted strategies to improve KMC coverage in low-resource settings. Kangaroo Mother Care (KMC) reduces neonatal mortality and improves thermoregulation and breastfeeding, yet uptake remains inconsistent in Sierra Leone. Predictive and explainable tools could target implementation where the need is most significant and resources are scarce. This study aimed to predict KMC adoption and identify actionable predictors using explainable machine learning. We analysed a nationally representative dataset from Sierra Leone comprising 7737 births. The study setting was Sierra Leone's healthcare system, with participants including mothers who delivered in health facilities. Following data preprocessing (imputation, MinMax normalisation, categorical encoding, and SMOTE for class imbalance), forward-backward selection reduced 22 candidate variables to 10 key predictors. Five classifiers were trained using a 70:30 stratified split: K-Nearest Neighbors (KNN), logistic regression (LR), Support Vector Machine (SVM), Random Forest (RF), and XGBoost. The outcome was KMC adoption (binary: received/not received). Performance was evaluated using accuracy, precision, recall, F1-score, and ROC-AUC. Interpretability was achieved through SHAP and LIME for global and local explanations. XGBoost performed best (accuracy 0.72, precision 0.75, recall 0.81, F1 0.78, ROC AUC 0.7685), followed by Random Forest. Predictors associated with KMC included delivery by caesarean section, type of birth, maternal employment, number of antenatal visits, place of delivery, health insurance coverage, and region, while sampling design variables captured contextual heterogeneity. SHAP and LIME consistently highlighted delivery characteristics and socio-economic factors as primary drivers. Explainable ensemble models can flag infants likely to receive or miss KMC and indicate modifiable levers for improvement. High recall supports use as a screening aid to prioritise counselling, facility preparedness, and postnatal support. Prospective validation, threshold calibration, and integration within routine health information systems are warranted to translate these insights into sustained increases in KMC coverage in Sierra Leone and similar settings.