Custom machine learning algorithm for large-scale disease screening - taking heart disease data as an example

Abstract

Heart disease accounts for millions of deaths worldwide annually, representing a major public health concern. Large-scale heart disease screening can yield significant benefits both in terms of lives saved and economic costs. In this study, we introduce a novel algorithm that trains a patient-specific machine learning model, aligning with the real-world demands of extensive disease screening. Customization is achieved by concentrating on three key aspects: data processing, neural network architecture, and loss function formulation. Our approach integrates individual patient data to bolster model accuracy, ensuring dependable disease detection.We assessed our models using two prominent heart disease datasets: the Cleveland dataset and the UC Irvine (UCI) combination dataset. Our models showcased notable results, achieving accuracy and recall rates beyond 95 % for the Cleveland dataset and surpassing 97 % accuracy for the UCI dataset. Moreover, in terms of medical ethics and operability, our approach outperformed traditional, general-purpose machine learning algorithms. Our algorithm provides a powerful tool for large-scale disease screening and has the potential to save lives and reduce the economic burden of heart disease.