Deep Learning Method to Estimate Glucose Level from Heart Rate Variability

Recently, there have been efforts by different researchers in implementing some Machine Learning (ML) and Deep Learning (DL) techniques in designing a model that will predict Glucose based solely on Heart Rate Variability parameters. However, each study uses an in-house dataset and thus the results differ from the rest. The aim of this paper is to explore the predictive capabilities of DL techniques in designing a model to predict glucose regulation from Heart Rate Variability. The clinical study was conducted on a dataset of 155 patients with long-term Electrocardiogram measurements. The best results are achieved with an architecture of three hidden layers (32, 256, and 64 neurons, respectively) with an Adam optimizer alongside a learning rate of 0.001 coupled by a Binary Cross entropy loss function. Furthermore, the Z score outlier removal method proved to lead to a higher accuracy value, whilst the IQR outlier removal method proved to lead to a higher Fl score value.