EFFICIENCY IN PREDICTING THE RISK OF DIABETES JOINTLY WITH THE RISK OF HYPERTENSION USING DEEP LEARNING FOR MULTI-LABEL CLASSIFICATION

Abstract

In the medical field, deep learning is widely used to create predictive models, which outperform traditional models. However, patients can suffer from multiple diseases simultaneously. To address this, deep learning has been developed to predict multiple diseases at once, known as multi-label classification neural networks. Although neural networks have shown great potential for prediction, they still face challenges with limited data. This research aims to improve the overall performance of the model by studying the relationship between labels using data on diabetes and hypertension, which often co-occur. The experiments were divided into two parts: simulated data and real-world data to compare multi-label and single-label feedforward neural networks. The results showed that multi-label neural networks performed well theoretically when tested on simulated data. However, in real-world data, using related labels did not significantly reduce the loss function but had the advantage of mitigating overfitting and maintaining comparable predictive performance to using a single label.