Heart disease prediction using rough neutrosophic sets and dual-attention neural networks: RNS-OptiDANet

Area of Science:

• Cardiology
• Artificial Intelligence
• Data Science

Background:

• Heart disease poses a significant global health challenge, necessitating advanced prediction methods.
• Current prediction models require enhanced accuracy and interpretability for clinical application.

Purpose of the Study:

• To introduce RNS-OptiDANet, a hybrid framework for accurate heart disease prediction.
• To leverage rough set theory (RST), rough neutrosophic sets (RNS), and an optimized dual-attention neural network (OptiDANet) for improved classification.
• To enhance model interpretability and transparency through an eXplainable Artificial Intelligence (XAI) module.

Main Methods:

• Feature selection using RST QuickReduct with discernibility matrix (RST QRDM).
• Representation of selected features using RNS to manage classification uncertainty.
• Implementation of OptiDANet with Channel Attention (CAM) and Soft Attention Mechanism (SAM) for pattern highlighting and noise reduction.
• Hyperparameter tuning with Optuna to optimize performance and prevent overfitting.
• Final classification using a Random Forest (RF) model.

Main Results:

• The RNS-OptiDANet framework demonstrated strong performance across datasets.
• Key performance metrics including accuracy, precision, recall, and F1-score were significantly improved.
• Experimental results confirmed the model's effectiveness in heart disease prediction.

Conclusions:

• The proposed RNS-OptiDANet model offers a robust and effective solution for heart disease prediction.
• The integrated XAI module provides crucial feature-level interpretability and clinical transparency.
• Ablation studies validated the significant contribution of each component, confirming the framework's overall efficacy.