In this work, we propose a structure-aware classifier (STAC) supported with a conditional tabular variational auto-encoder (TVAE) to predict cardiovascular disease. A K-Nearest Neighbours Graph (KNN Graph) is deduced from the clustering of the class samples and the distances to edge segments are calculated. These differences are then sorted and first n values are summed. Depending on this sum, a prediction is made for heart disease. Backed with a Standard Scaler and a conditional TVAE, the classifier achieves 0.977 accuracy on Cleveland dataset while the state-of-the-art models remain at 0.884 accuracy. Ablation studies are also performed to measure the impact of each component in the system. Additional variation experiments are conducted to compare the structure-aware classifier with traditional classifiers such as Linear Discriminant Analysis (LDA) and Gradient Boosting Machines (GBM). We also plugged the General Performance Score (GPS) metric to fuse the metrics such as Accuracy, Precision, F1-Score, Sensitivity, and Specificity. Results indicate that the proposed method has better prediction performance than the classical methods.