Enhancing Clinical Decision-Making: Comparative Evaluation of Machine Learning Models for Symptom-Based Disease Diagnosis

Healthcare faces a tough challenge with misdiagnosis, which contributes to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{1 0 - 1 5}$</tex> percent of adverse medical events globally, according to the World Health Organization. Accurately identifying diseases based on symptoms is challenging because many conditions share similar signs, and traditional methods often struggle to keep up. In this study, we explored the Symptom-Disease Prediction Dataset, which contains 4961 patient records, to develop a machine learning solution for better diagnosis. We tested various models, including Random Forests, Support Vector Machines and Neural Networks, when combined using ensemble techniques. Our K-Nearest Neighbors model proved to be the standout, achieving a solid accuracy of 90.74 percent, surpassing other approaches, with a precision of 90.37 percent and an F1-score of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{8 9. 3 2 \%}$</tex>. This study offers a useful solution for improving diagnosis, minimizing mistakes, and building trust in AI-powered healthcare tools. We suggest conducting experiments in real-world healthcare settings to enable the use of this method on a daily basis and improve patient care.