Advancing Cardiovascular Disease Network Analysis: A Comparative Study of Correlation-Based Graph Construction and Link Prediction

Cardiovascular diseases (CVDs) are a leading cause of mortality worldwide, and accurate predictive models are essential for improving prevention and management strategies. This study addresses the challenge of enhancing CVD risk prediction through correlation-based graph construction and weighted link prediction algorithms. Using Pearson and Spearman correlation methods, we transformed a comprehensive dataset containing 1025 patient records and 14 key features into graph structures. Correlationbased graph construction captures feature dependencies by representing variable relationships as edges in a network. To evaluate the effectiveness of the graph representations, we applied weighted link prediction algorithms, including Weighted Common Neighbors (WCN), Weighted Preferential Attachment (WPA), and Weighted Jaccard Coefficient (WJC). The Pearson correlation-based network demonstrated exceptional performance, with the WCN algorithm achieving an Area Under the Curve (AUC) of 99.80% and a Precision of 48.0%. In contrast, the Spearman correlation-based network showed robust results, with WJC achieving an AUC of 96.60% and Precision of 67.16%. The comparative analysis, conducted using Python in a Jupyter environment and employing libraries such as NetworkX and various statistical libraries, highlights the superior ability of correlation-based graphs to capture linear and non-linear relationships in CVD data. While promising, the study acknowledges limitations related to dataset size and computational complexity. Our findings suggest that correlationbased graph methods significantly enhance CVD prediction, offering a more personalized CVD prevention and management approach.