Electrocardiographic and Biochemical Feature Integration for Automated Cardiovascular Risk Stratification

This work explored how machine learning can help identify patients with Congestive Heart Failure (CHF) using both ECG readings and biochemical test results. The dataset included 1,000 patient records with structured features from lab reports, ECG intervals, clinical signs, and diagnostic history. After cleaning and balancing, four models were trained: Logistic Regression, Random Forest, XGBoost, and a neural network. Accuracy was high across the board, but most models failed to detect the CHF-positive cases reliably. Some, like XGBoost, did not identify a single case. The neural model performed better once its decision threshold was adjusted. At a threshold of 0.3, it reached a recall of 0.18 and an F1-score of 0.19 for the CHF class, better than any other model tested. These results are not final, and the model will need to be tested on broader clinical data. But they suggest that simple changes like threshold tuning can help machine learning systems catch more high-risk cases without needing major redesign.