PBCC-SFOpt1DCNN: A Starfish Optimization-Enhanced 1DCNN Model for Heart Disease Prediction with PBCC-Based Feature Selection

Early and accurate prediction is essential for the prevention of heart disease, a global health concern with a high fatality rate. For quick and accurate diagnosis, machine learning (ML) and deep learning (DL) have shown promise. However, issues including overfitting, inefficiency, and poor prediction accuracy are frequently encountered with conventional ML techniques. This study offers a novel method to improve the accuracy of HD prediction by applying a combined intelligent system in order to overcome these constraints. Three benchmark datasets for heart illness from the Kaggle repository have been used for tests and evaluations. In order to improve dataset quality and avoid distortions, our approach starts with the pre-processing step, where four data preparation phases are used: normalization, data encoding, handling imbalanced data, and data splitting. Feature selection is performed using a Point Biserial Correlation Coefficient (PBCC) method. Finally, a Starfish Optimization-based 1DCNN is proposed for accurate HD prediction, in which one-dimensional convolutional neural network hyper-parameters are tuned by SFOA, enhancing model performance, termed SFOpt1DCNN. The findings show that the Hybrid PBCC-SFOpt1DCNN obtained over 98% accuracy on the Framingham HD as well as the Indicators of HD datasets and the Cleveland HD dataset. By offering a robust and efficient framework for HD prediction, the Hybrid PBCC-SFOpt1DCNN method helps radiologists and doctors make more precise diagnoses.