Adaptive Hybrid Cardiovascular Disease Prediction Using Deep Neural Networks and Anomaly-Aware Risk Scoring

Abstract

Cardiovascular disease (CVD) is a major health concern, making early risk identification an important task in healthcare decision support. This paper presents an adaptive hybrid cardiovascular disease prediction framework that combines a Deep Neural Network (DNN), Isolation Forest-based anomaly detection, and SHAP (SHapley Additive exPlanations) for interpretable risk assessment. The system uses clinical attributes from the UCI Cleveland Heart Disease dataset. Categorical features are transformed using one-hot encoding and the resulting features are scaled using StandardScaler. The DNN learns nonlinear relationships among clinical attributes and estimates the probability of cardiovascular disease, while the Isolation Forest identifies unusual patient patterns. These outputs are integrated using a hybrid risk-scoring mechanism in which the disease probability contributes 80% and the normalized anomaly component contributes 20%. SHAP is used to identify the features that influence individual predictions. The complete framework is implemented as an interactive Streamlit dashboard that presents disease probability, anomaly score, hybrid risk score, risk category, and feature-level explanations. An automated PDF reporting module additionally summarizes the prediction and clinical interpretation. The proposed framework therefore integrates prediction, anomaly awareness, explainability, and automated reporting into a unified cardiovascular disease risk-assessment system.

Introduction

This text presents a hybrid, interpretable cardiovascular disease prediction system designed to go beyond simple disease classification. The system combines three main AI techniques:

• Deep Neural Network (DNN): Predicts the probability that a patient has cardiovascular disease based on clinical features.

• Isolation Forest: Detects unusual or atypical combinations of patient characteristics that may not be captured by disease prediction alone.

• SHAP Explainable AI: Explains how individual features contribute positively or negatively to a particular prediction.

The system uses a UCI heart disease dataset containing 920 records and 16 original columns. Clinical data are preprocessed through missing-value handling, one-hot encoding of categorical variables, and StandardScaler-based numerical scaling. The resulting 18-feature input vector is passed to both the DNN and Isolation Forest.

The outputs are combined into an anomaly-aware Hybrid Risk Score:

H = 0.80 × DNN Probability + 0.20 × Normalized Anomaly Score

The resulting score is categorized as Low Risk (0–0.30), Moderate Risk (0.30–0.70), or High Risk (0.70–1.00). The 80:20 weighting is an implementation choice rather than a clinically optimized weighting.

For evaluation, the system used a held-out test set of 184 samples, with 67 true negatives, 15 false positives, 12 false negatives, and 90 true positives. The reported ROC-AUC is 0.913. Some original training details, such as the exact train/test split, random seed, cross-validation method, and complete hyperparameters, are unavailable from the current artifacts and therefore are not assumed.

The research gap addressed by the project is that many existing cardiovascular prediction systems focus primarily on classification and predictive accuracy. The proposed framework instead integrates prediction, anomaly detection, risk scoring, explainability, and reporting into one workflow.

Finally, the results are presented through an interactive Streamlit dashboard and can be exported as a PDF report, providing users with the predicted risk, anomaly information, risk category, and SHAP-based explanation of the prediction.

In one sentence

The project develops an interpretable hybrid cardiovascular risk-assessment system that combines DNN disease prediction, Isolation Forest anomaly detection, SHAP explanations, and a hybrid risk score within an interactive dashboard and reporting framework.

Conclusion

This paper presents an adaptive hybrid cardiovascular disease prediction framework that integrates Deep Neural Networks, Isolation Forest anomaly detection, hybrid risk scoring, and SHAP Explainable AI. The DNN provides disease probability estimation, while the Isolation Forest adds anomaly awareness to the prediction workflow. The outputs are combined through a hybrid risk score and categorized into low, moderate, and high-risk levels. The integration of SHAP provides feature-level explanations, improving the interpretability of the prediction. Furthermore, the Streamlit dashboard and automated PDF reporting module transform the machine-learning pipeline into an interactive application capable of presenting prediction results and explanations in a user-friendly format. The implemented framework therefore provides an integrated approach combining prediction, anomaly detection, explainability, and reporting for cardiovascular disease risk assessment.

References

[1] Krittanawong, C., Virk, H. U. H., Bangalore, S., Wang, Z., Johnson, K. W., Pinotti, R., Zhang, H., Kaplin, S., Narasimhan, B., Kitai, T., Baber, U., Halperin, J. L., & Tang, W. H. W. (2020). Machine learning prediction in cardiovascular diseases: A meta-analysis. Scientific Reports, 10(1), Article 16057. [2] Rahim, A., Rasheed, Y., Azam, F., Anwar, M. W., Rahim, M. A., & Muzaffar, A. W. (2021). An integrated machine learning framework for effective prediction of cardiovascular diseases. IEEE Access, 9, 106575–106588. [3] Ghosh, P., Azam, S., Jonkman, M., Karim, A., Shamrat, F. M. J. M., Ignatious, E., Shultana, S., Beeravolu, A. R., & de Boer, F. (2021). Efficient prediction of cardiovascular disease using machine learning algorithms with Relief and LASSO feature selection techniques. IEEE Access, 9, 19304–19326. [4] Li, Q., Campan, A., Ren, A., & Eid, W. E. (2022). Automating and improving cardiovascular disease prediction using machine learning and EMR data features from a regional healthcare system. International Journal of Medical Informatics, 163, Article 104786. [5] Rahmathullah, M., Bhavanam, S. N., & Midasala, V. (2025). Deep learning-powered cardiovascular disease prediction: A novel approach to early diagnosis and risk assessment. Journal of Neonatal Surgery, 14(4), 21–31.

Copyright

Copyright © 2026 Golagani Usha Sri, Dr. Golagani Lavanya Devi. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.