A mixed time series glycemic prediction model

The article considers the problem of predicting blood glucose levels using data containing weakly expressed dependencies between parameters, including time series and physiological parameters. An approach based on the use of neural networks with long short-term memory (LSTM) is proposed, which is capable of predicting future glucose values (SGV), as well as identifying anomalies in the data. To improve the quality of the model, the DBSCAN clustering method is used, which allows you to identify groups of data with similar characteristics. An algorithm for filling in missing data based on the average value in the cluster is also developed, which improves the accuracy of forecasting. Numerical experiments were carried out on data collected by monitoring glucose levels, which demonstrated the effectiveness of the proposed approach for predicting SGV, taking into account time dependencies and the influence of associated factors.