Leila Aissaoui ^1* Manel Ben amar ²

• ¹ Ecole Supérieure des Communications de Tunis, Université de Carthage, Ariana, Tunisia
• ² Département de Médecine Communautaire, Faculté de Médecine de Monastir, Université de Monastir, Monastir, Monastir, Tunisia

This study presents the development and evaluation of machine learning models for a symptom-based health checker using a dataset comprising 10 diseases and 9572 samples. The dataset was divided into training and testing sets to facilitate model training and evaluation. Decision Tree, Random Forest, Naive Bayes, Logistic Regression, and K-Nearest Neighbors models were selected and optimized to maximize performance. Evaluation metrics including accuracy, F1 scores, and 10-fold cross-validation were employed to assess model performance. Clinical vignettes were utilized to gauge the real-world applicability of the models, demonstrating their robustness in providing accurate diagnoses. Additionally, the role of ROC-AUC curves in assessing model performance was explored, revealing improvements in model performance with increasing complexity. Precision-recall curves were also instrumental in evaluating model sensitivity, particularly in scenarios with imbalanced datasets. Overall, this study underscores the importance of comprehensive model evaluation techniques, including clinical vignette testing and analysis of ROC-AUC and precision-recall curves, in ensuring the reliability and sensitivity of symptom-based health checkers.