<![CDATA[Hypertension is one of the leading causes of premature death globally that often goes undetected due to minimal clinical symptoms, earning it the nickname “silent killer.” The application of artificial intelligence (AI), particularly Machine Learning, is a strategic approach to early detection, but the main challenge lies in balancing diagnostic accuracy with detection sensitivity so that no patients at risk are overlooked. This study aims to analyze and compare the performance of the Extreme Gradient Boosting (XGBoost) algorithm with the Cost-Sensitive strategy compared to Gaussian Naive Bayes (GNB) as a baseline in hypertension risk classification. The dataset used included 1,985 electronic medical records with 9 clinical attributes, which were evaluated using the 10-Fold Cross-Validation method to determine model validity. The test results showed that XGBoost consistently outperformed GNB across all evaluation metrics. XGBoost recorded superior performance with an Accuracy of 92.19% and an AUC of 0.9752, far surpassing GNB, which obtained an Accuracy of 84.13%. The application of Cost-Sensitive Learning in XGBoost proved effective in overcoming performance trade-offs by producing a Recall of 91.26% and a Precision of 93.53%. Furthermore, Feature Importance analysis identified Blood Pressure History, Smoking Status, and Family History as the most dominant risk factors, which is in line with global medical guidelines. Based on these results, it is concluded that XGBoost is a more reliable and accurate method to be applied in early detection systems for hypertension compared to classical probabilistic approaches.]]>
