Development of an Artificial Intelligence I-Based Symptom Checker for Rural Clinics

Access to timely and accurate healthcare in rural areas is often limited by the scarcity of trained medical professionals and diagnostic facilities. This study presents the design, implementation, and comprehensive evaluation of an AI‑Based Symptom Checker tailored for rural clinics, aiming to support healthcare workers in making informed, evidence‑based decisions. The system combines an XGBoost machine learning classifier with SHAP (SHapley Additive exPlanations) to provide accurate disease diagnoses and transparent explanations of the factors influencing each prediction. The training dataset comprised 8,247 patient records (53.1% from local rural clinics in Enugu State, Nigeria, and 46.9% from MIMIC‑IV and WHO repositories) spanning 23 disease classes. Rigorous evaluation through 10‑fold cross‑validation yielded a mean accuracy of 91.2% (95% CI: 91.0‑91.4%), precision of 90.6% (95% CI: 90.4‑90.8%), recall of 91.0% (95% CI: 90.8‑91.2%), F1‑score of 90.8% (95% CI: 90.6‑91.0%), and AUC‑ROC of 0.94 (95% CI: 0.939‑0.941). Comparative benchmarking confirmed that XGBoost significantly outperformed Logistic Regression, Random Forest, SVM, and Neural Network baselines (p < 0.05 for all comparisons). SHAP explanations achieved a mean usability score of 4.22/5.0 in structured evaluation with 20 healthcare workers, with 89.4% clinician agreement on feature relevance. Ethical approval, patient consent frameworks, NDPR‑compliant data governance, and clinical risk mitigation strategies were established. The system demonstrates that an AI‑Based Symptom Checker is a viable, accurate, and interpretable decision support tool that can improve diagnostic quality, enhance triage care, and offer significant benefits to patients in resource‑constrained rural clinic settings.