A Comparative Study of Random Forest and K-Nearest Neighbors for Predicting Hypertension and Diabetes

Hypertension (HTN) and Diabetes Mellitus (DM) are global health threats with an increase in prevalence and economic burden. This study assessed how well the K-nearest neighbor (KNN) and Random Forest (RF) models performed, for predicting these diseases using a Kaggle dataset containing $\mathbf{2 6, 0 8 4}$ patient records. Models were trained on eight clinical variables, including age, cholesterol, and BP after preprocessing and principal component analysis (PCA) feature selection. The RF model performed well in predicting hypertension, demonstrating its robustness to class imbalance and ability to capture complex feature interactions, with 99.6 % accuracy, 100 % precision, 99.0 % recall, and only two false negatives. In contrast, KNN showed moderate results in predicting diabetes (82.95 % accuracy and 81.55 % precision) but with a high false negative rate (29.74 %), highlighting its sensitivity to noise and imbalance. These results make RF superior for screening hypertension, where minimizing underdiagnosis is critical. For predicting diabetes, the potential of KNN needs to be optimized using techniques such as SMOTE oversampling and parameter tuning. This study supports the use of radiofrequency technology in clinical settings. At the same time, further research is needed to improve the reliability of KNN. In future work, better and higher quality data standards can be incorporated, and these methods can be validated in medical practice to maximize their impact on the early intervention of disease.