Enhancing medical diagnosis with a machine learning-based symptom checker for health assessment

Abstract

The growing incidences of chronic illnesses and increased complexity of medical diagnoses point to the desire to have effective, precise, and timely diagnostic instruments to assist medical personnel. The goal of the current study is to develop an ML-based system that will be able to classify the symptoms and predict the diseases to reduce the delays and inconsistencies in the traditional diagnosis. To enhance the accuracy of prediction, bootstrapping and majority voting are used. Bootstrapping enables diverse training of multiple decision trees, while majority voting aggregates their predictions, reducing individual model bias and contributing to high predictive accuracy. It has been shown that as an experimental model, the model attains an accuracy of 98.33, a precision of 92.62%, a recall of 91.67%, and an F1-score of 91.58%. Also, ROC and Precision-Recall curves gave an AUC value nearer to 1.0, thus confirming high predictive accuracy. Even though the False Negative Rate (0.0806) was a little higher than the False Positive Rate (0.0161), the system is useful in the prediction of diseases and classification of symptoms, and future enhancements will be based on the hyperparameter tuning, incorporation of other algorithms, and real-time data through wearable sensors to improve the clinical applicability.