Database Proposal for Correlation of Glucose and Photoplethysmography Signals

This work presents a Database that contains Photoplethysmography signals, glucose levels, weight, height and age of 217 patients. The information of biologic activity was obtained using the handle Empatica E4 Wristband, the glucose level using laboratory blood chemistry analyzers (Cobas 6000), and the physical parameters using standardized instruments. The database comprises a forward training a total of 5576 samples and another segment of validation to a total of 2164 samples. The Database has been used to evaluate different prediction techniques based on Machine Learning (Random Forest, Artificial Neural Network, Support Vector Machine, Gradient Boosting Machine). The implementation of these algorithms provides up to 90% average accuracy, a correlation of 0.88 and a satisfactory evaluation in the Error Diagram of Clarke. According to the results obtained, the proposed database is appropriate for training and verification of existing correlation between photoplethysmography signals and blood glucose level.