A CNN-GRU framework for stroke-heart attack prediction using IMOWPA-tuned SMOTE and LZMA compression.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Data Science

Background:

• Healthcare datasets, particularly from intensive care units, often suffer from class imbalance, with conditions like stroke being underrepresented.
• Handling unbalanced data is a significant challenge in data mining and predictive modeling within healthcare.

Purpose of the Study:

• To develop a method for accurately identifying and categorizing minority-class data in highly imbalanced healthcare datasets.
• To predict stroke incidence using balanced and compressed data from the MIMIC III dataset.

Main Methods:

• A novel Convolutional Neural Network-Gated Recurrent Unit with Imbalanced Data Handling (CNN-GRU-IDH) model was proposed.
• Healthcare data was compressed using the Lempel Ziv Markov Chain Algorithm (LZMA) to reduce transfer volume.
• The Synthetic Minority Over-sampling Technique (SMOTE) was employed to address class imbalance, with its K nearest neighbor value optimized by the Improved Multi-Objective Wolf Pack Algorithm (IMOWPA).

Main Results:

• The proposed CNN-GRU-IDH model achieved an accuracy of 87.66% and an F1 score of 85.63% on a 70% training and 30% testing split.
• The model demonstrated superior performance compared to existing methods on the imbalanced MIMIC III dataset.
• The study successfully predicted stroke incidence using the developed classification technique.

Conclusions:

• The CNN-GRU-IDH approach offers a significant advancement in patient-specific early stroke prediction.
• This method has the potential to save lives and reduce mortality rates through improved early detection.
• The integration of data compression and advanced imbalance handling techniques enhances the reliability of predictive models in critical care scenarios.