Development of a Machine Learning-Based Symptom Checker for Early Disease Diagnosis

Abstract

Early identification of disease is critical for successful treatment, and this study introduces a machine learning-based symptom checker that use a Random Forest Classifier to improve diagnostic accuracy and reliability. The model, which was trained on a dataset of $\mathbf{4, 9 6 1}$ patient records containing 133 symptoms and 41 illnesses, uses Principal Component Analysis (PCA) to reduce dimensionality, assuring computational efficiency while keeping crucial diagnostic trends. An 80-20 train-test split is utilized, with hyperparameter adjustment to avoid overfitting. Random Forest outperforms Decision Tree, K-Nearest Neighbors, and Naïve Bayes classifiers in handling unbalanced medical data, leading to enhanced interpretability. The model’s robustness, which achieved $93.25 \%$ accuracy, an $F 1$-score of 0.9150, and a geometric mean of 0.96, is further confirmed using precision, recall, and the Matthews correlation coefficient. This work addresses the potential of AI-driven decision-support systems to revolutionize healthcare analytics and provide accurate, accessible, and automated illness prediction. It does this by addressing diagnostic discrepancies and enabling scale implementation in telemedicine.

Ähnliche Arbeiten