Next-Gen Heart Disease Prediction: A Gradient Boosting Approach with L1-Regularized Neural Networks

Heart diseases, also known as diseases caused by cardiovascular complexities, cover a range of heart and blood vessels affecting condition, emerging as a notable concern in global health. Timely detection of any cardiovascular disease is of immense significance to effective treatment and prevention. This paper introduces a new method to diagnose and predict cardiac disease using gradient-boosting classifier with L1-regularized neural networks as a base estimator. The proposed method can effectively handle small datasets and resist overfitting, a common challenge in traditional machine-learning methods. Using a neural network as the base estimator of a gradient-boosting Classifier allows the model to capture nonlinear relationships between risk factors and improve prediction accuracy. Besides, L1 regularization, in particular, can effectively promote sparsity in the neural network weights, which can be considered as helpful to identify the most important predictors of heart disease. Five publicly accessible cardiac disease datasets are used to evaluate the suggested methodology. Experimental results demonstrating better performance of the proposed strategy than conventional machine learning approaches, achieving 97.11% accuracy, 95.51% precision, 98.84% recall, and 97.14% F1-score. The suggested approach can deliver precise forecasts, and is robust to noise and outliers, making it a valuable technique for clinical diagnosis and decision-making.