GPCB: A Hybrid GA-PSO and Transformer-guided CNN-BiLSTM Framework for Cardiovascular Disease Prediction

Cardiovascular disease (CVD) remains the leading cause of death worldwide, so predicting risk as soon as possible and accurately is necessary to provide timely intervention.However, previous interpretive models perform poorly due to overfitting, poor feature selection, and substantively poor management of heterogeneous clinical and behavioral data.To improve the classification of CVD risk, this paper presents a framework termed as GPCB.This combines Genetic Algorithm -Particle Swarm Optimization (GA-PSO) for feature selection with a deep learning model construction that includes transformers, Convolutional Neural Networks (CNN), and Bidirectional Long Short Term Memory (Bi-LSTM) models (T-CBLSTM).During phase I, the GA-PSO module performs multi-objective feature optimization when predicting by comparing and assessing predictive relevance and minimizing input dimensions, allowing a reasonable selection of clinical and lifestyle features that were meaningfully relevant.In phase II, a feature extracted and selected T-CBLSTM model was constructed to compose the model, where the CNN layers extracted spatial patterns, the Transformer blocks accounted for global dependencies in the data, and the Bi-LSTM layers attended to the sequential relationships.This framework was evaluated on the UCI Heart Disease, Framingham Heart Study, and MIMIC-III datasets as well as the merged datasets.The experimental results demonstrate that GPCB-TC outperformed the state-of-the-art accuracy up to 98.3%, F1-score 97.6%, and AUC-ROC 0.98.The proposed model shows immense opportunities for development and implementation in clinical decision support systems by offering risk assessment in a real-world healthcare practice.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.