Enhancing Typhoid Fever Diagnosis Based on Clinical Data Using a Lightweight Machine Learning Metamodel

Background: Typhoid fever remains a significant public health challenge, especially in developing countries where diagnostic resources are limited. Accurate and timely diagnosis is crucial for effective treatment and disease containment. Traditional diagnostic methods, while effective, can be time-consuming and resource-intensive. This study aims to develop a lightweight machine learning-based diagnostic tool for the early and efficient detection of typhoid fever using clinical data. Methods: A custom dataset comprising 14 clinical and demographic parameters-including age, gender, headache, muscle pain, nausea, diarrhea, cough, fever range (°F), hemoglobin (g/dL), platelet count, urine culture bacteria, calcium (mg/dL), and potassium (mg/dL)-was analyzed. A machine learning metamodel, integrating Support Vector Machine (SVM), Gaussian Naive Bayes (GNB), and Decision Tree classifiers with a Light Gradient Boosting Machine (LGBM), was trained and evaluated using k-fold cross-validation. Performance was assessed using precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC). Results: The proposed metamodel demonstrated superior diagnostic performance, achieving a precision of 99%, recall of 100%, and an AUC of 1.00. It outperformed traditional diagnostic methods and other standalone machine learning algorithms, offering high accuracy and generalizability. Conclusions: The lightweight machine learning metamodel provides a cost-effective, non-invasive, and rapid diagnostic alternative for typhoid fever, particularly suited for resource-limited settings. Its reliance on accessible clinical parameters ensures practical applicability and scalability, potentially improving patient outcomes and aiding in disease control. Future work will focus on broader validation and integration into clinical workflows to further enhance its utility.