Optimized robust learning framework based on big data for forecasting cardiovascular crises.

Area of Science:

• Healthcare analytics
• Artificial intelligence in medicine
• Cardiovascular disease prediction

Background:

• Deep learning (DL) in healthcare requires robust pre-processing for noisy datasets.
• Existing frameworks struggle with efficiency and data handling.
• Cardiovascular crisis prediction is critical for timely intervention.

Purpose of the Study:

• To propose a novel robust deep learning framework (R-DLH2O) for cardiovascular crisis forecasting.
• To enhance prediction accuracy and efficiency through a multi-phase approach.
• To integrate advanced feature selection and data pre-processing techniques.

Main Methods:

• Developed the R-DLH2O framework with five phases: robust pre-processing, feature selection, feed-forward neural network, prediction, and evaluation.
• Utilized H2O for big data processing.
• Introduced a Modified Whale Optimization Algorithm (MWOA) with Gaussian distribution for random walks and diffusion strategy.

Main Results:

• The R-DLH2O framework achieved 95.93% accuracy, 92.57% precision, and 93.6% recall.
• Processing time was 436 seconds with a mean per-class error of 0.150125.
• The Modified WOA (MWOA) demonstrated superior accuracy and robustness over the standard WOA.

Conclusions:

• The R-DLH2O framework offers a significant advancement in cardiovascular crisis prediction.
• The proposed MWOA enhances the performance of deep learning models in healthcare.
• The framework provides robust and efficient healthcare analytics, outperforming previous methods.