Optimizing stroke prediction using gated recurrent unit and feature selection in Sub-Saharan Africa.

Area of Science:

• Neurology
• Artificial Intelligence
• Public Health

Background:

• Stroke is a major global health issue, disproportionately affecting African populations due to healthcare disparities.
• Early stroke prediction and intervention are crucial for improving patient outcomes.
• This research addresses the need for advanced predictive tools in stroke management.

Purpose of the Study:

• To develop and evaluate a novel stroke prediction system utilizing Gated Recurrent Units (GRU).
• To leverage the Afrocentric Stroke Investigative Research and Education Network (SIREN) dataset for model training and validation.
• To compare the GRU model's performance against traditional machine learning algorithms and Long Short-Term Memory (LSTM) networks.

Main Methods:

• Utilized secondary data from the SIREN dataset (4236 records, 29 phenotypes).
• Applied feature selection to identify 15 optimal phenotypes for stroke prediction.
• Trained a GRU model with specific architecture and hyperparameters, evaluated using accuracy, AUC, and prediction time.

Main Results:

• The GRU-based system achieved 77.48% accuracy and an AUC of 0.84.
• GRU model demonstrated a rapid prediction time of 0.43 seconds, outperforming LSTM (2.23 seconds).
• Feature selection significantly enhanced model performance compared to using all available phenotypes.

Conclusions:

• The GRU model offers a superior, efficient, and scalable solution for stroke prediction.
• Future work should involve integrating diverse data types and validating on varied populations.
• Exploring hybrid AI architectures could further improve predictive capabilities for stroke.