A Hybrid CNN–BiLSTM Frame work for Heart Disease Detection

Resumo

Heart disease continues to rank among the primary causes of death globally, emphasizing the need for precise and user-friendly prediction algorithms. We propose a novel hybrid deep learning model in this study for forecasting heart disease from structured clinical data by combining CNNs along Bidirectional Long Short-Term Memory (BiLSTM) networks. Second-degree polynomial feature expansion or normalization for numerical stability is used to improve the model’s capacity to represent intricate relationships. We also use the Synthetic Minority Oversampling Technique (SMOTE) to handle class imbalance and reformat tabular data into a pseudo-sequential style in order to take advantage of sequence modeling. Our CNN–BiLSTM model achieves 98% accuracy and a 0.98 F1-score, which is a considerable improvement over the baseline machine learning classifiers, such as Decision Trees, Naive Bayes, and Logistic Regression. These results demonstrate how beneficial it is to combine local pattern extraction with temporal modeling to obtain more accurate disease prediction.