Physical Activity Integration in Blood Glucose Level Prediction: Different Levels of Data Fusion.

Abstract

Blood glucose level (BGL) prediction contributes to more effective management of type 1 diabetes. Physical activity (PA) is a crucial factor in diabetes management. It affects BGL, and it is imperative to effectively deploy PA in BGL prediction to support diabetes management systems by incorporating this crucial factor. Due to the erratic nature of PA's impact on BGL inter- and intra-patients and insufficient knowledge, deploying PA in BGL prediction is challenging. Hence, optimal approaches for PA fusion with BGL are demanded to improve the performance of BGL prediction. To address this gap, we propose novel methodologies for extracting and integrating information from PA data into BGL prediction. This paper proposes several novel PA-informed prediction models by developing different approaches for extracting information from PA data and fusing this information with BGL data in signal, feature, and decision levels to find the optimal approach for deploying PA in BGL prediction models. For signal-level fusion, different automatically-recorded PA data are fused with BGL data. Also, three feature engineering approaches are developed for feature-level fusion: subjective assessments of PA, objective assessments of PA, and statistics of PA. Furthermore, in decision-level fusion, ensemble learning is used to combine predictions from models trained with different inputs. Then, a comparative investigation is performed between the developed PA-informed approaches and the no-fusion approach, as well as between themselves. The analyses are performed on the publicly available Ohio dataset with rigorous evaluation. The results show that deploying PA can statistically significantly improve BGL prediction performance. The results show that deploying PA can statistically significantly improve BGL prediction performance. Also, among the developed approaches to leveraging PA in BGL prediction, fusing heart rate data at the signal-level and PA intensity categories at the feature-level with BGL data are the most effective ways. Our developed methodologies contribute to determining optimal approaches, including the kind of PA information and fusion method, to improve the performance of BGL prediction effectively.