Overview

The performance of the Stacking Gen AI model on the 400,000-record dataset, as shown in the provided table and ROC plot, can be summarized as follows:

Dataset	Performance	Model									Proposed Model
		LR	RF	GBM	XGB	CNN	GRU w/ Attention	CNN w/ GRU	Stacking ML+DL	Gen AI	Stacking Gen AI
400,000	Accuracy	77	90	77	80	78	79	80	90	95	96
	ROC AUC	84	96	85	88	86	87	88	96	99	99

1. Accuracy: The Stacking Gen AI model achieves a top-tier accuracy of 96%, which is the highest among all listed models. This high accuracy indicates that the model is highly effective in making correct predictions on the dataset compared to other machine learning and deep learning models, such as Random Forest (90%), Gradient Boosting (77%), and even the baseline Stacking model (90%).

1. ROC AUC Score: The model also reaches a near-perfect ROC AUC of 0.99, tying with the Gen AI Model. This high ROC AUC value means the Stacking Gen AI model has an excellent ability to distinguish between the positive and negative classes across various threshold settings. A ROC AUC of 0.99 implies that the model can achieve high true positive rates while keeping false positive rates low, which is critical in medical predictions like heart failure.
2. Comparison to Other Models:
   • The Logistic Regression model has a significantly lower ROC AUC (0.84) and accuracy (77%), demonstrating the limitation of simpler, linear models on complex datasets.
   • Random Forest (RF) performs relatively well, with an ROC AUC of 0.96 and an accuracy of 90%, but it is still outperformed by the Stacking Gen AI model. This indicates that the integration of multiple models in Stacking Gen AI provides added predictive strength.
   • Deep learning models like CNN with GRU and GRU with Attention show ROC AUC scores of 0.88 and 0.87, respectively. While these models perform well, they don’t reach the effectiveness of the Stacking Gen AI model, likely due to the additional power of GAN-generated synthetic data for class balancing in the Stacking Gen AI approach.
3. Impact of GAN Integration:
   • The GAN-generated data within the Stacking Gen AI model addresses class imbalance, improving minority class detection. This is evident in the high ROC AUC score, which reflects the model’s balanced performance across both majority and minority classes, a common challenge in medical datasets.
4. Overall Performance on the ROC Curve:
   • On the ROC curve, the Stacking Gen AI model (indicated by the bright green line) consistently remains near the top-left corner, confirming its high sensitivity and specificity. This positioning visually reinforces the model’s strong predictive capability and minimal compromise between true positives and false positives.

Conclusion

The findings in this study position the Stacking Generative AI model as a significant hybrid approach in predictive modeling for heart disease. By integrating Generative AI techniques with traditional machine learning models such as Random Forest (RF), Gradient Boosting Machines (GBM), and Extreme Gradient Boosting (xGBM), alongside deep learning architectures such as Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN), the model effectively combines the strengths of these methods to achieve superior predictive accuracy. The inclusion of Generative AI, particularly Generative Adversarial Networks (GANs), enhances the model’s ability to perform data augmentation and address class imbalances, which are common and critical challenges in healthcare datasets.

The model consistently demonstrated high accuracy and ROC AUC scores across datasets ranging from 303 to 400,000 records; it has been implemented and proved in my comprehensive research for the dissertation. For smaller datasets, such as the 303-record dataset, it achieved an ROC AUC of 0.99, outperforming standalone models like RF, CNN, and xGBM. On larger datasets, such as the 400,000-record dataset, the model maintained its strong performance with an accuracy of 96% and an ROC AUC of 0.99. These results confirm the model’s scalability and generalizability, making it highly effective for small-scale and large-scale clinical applications.

A key strength of this approach lies in the integration of Generative AI within the stacking framework. The use of GANs enables the generation of synthetic data to balance underrepresented classes, ensuring more robust training and improving the model’s ability to generalize across diverse and imbalanced datasets. This feature is particularly advantageous in healthcare, where imbalanced datasets are prevalent, and accurate predictions for minority classes, such as high-risk patients, are essential for improving clinical outcomes.

The Stacking Generative AI model’s hybrid structure leverages the strengths of traditional machine learning and deep learning methods. While traditional models like RF and xGBM excel in analyzing structured data and assessing feature importance, deep learning models like CNNs and RNNs effectively capture nonlinear relationships and temporal patterns. The seamless integration of these approaches allows the Stacking Generative AI model to provide a comprehensive solution for heart disease prediction.

Overall, these findings underscore the practical relevance of the Stacking Generative AI model for healthcare applications. Its adaptability to diverse dataset sizes and structures and consistent high performance make it a valuable tool for clinical decision-making. The model’s ability to improve predictive accuracy and address key challenges in healthcare data highlights its potential to advance heart disease prediction and management. Furthermore, this work lays the groundwork for future research exploring the broader application of Generative AI and hybrid models in medical prediction tasks.

LEARN MORE FROM THIS APPLICATION