Deep learning-based data imputation on time-variant data using recurrent neural network

Abstract

In general, numerous inbuilt diagnosis complications are due to improper or missing data. Thus, it becomes mandatory to perform proper imputation of the missed values to predict the diseases accurately. Imputation operations will be crucial when we encounter incompletely recorded patient data. The measurement of blood glucose level is considered to be the most important health-conscious effort that one does periodically since the false diagnosis of it leads to misinterpretation of patient health conditions that might cause fatal outcomes. But predicting those measures has become a tedious task in the course of diabetic treatment of these days. This paper focuses on the aim of the imputation of the missing patient-specific diabetic data, especially to overcome the existing methods’ demerits of yielding lesser accuracy and more time. This work attempts to predict the blood glucose levels by analyzing time-series data along with the patient activities. The patient activities are being thoroughly investigated here in this work; for instance, with the first 20-day diabetic data of a patient, the diabetic forecast for the next 10 days is made in the considered month. This prediction of patient diabetic conditions is done by proposing a novel approach for predicting the blood glucose levels with the aid of Maclaurin series-based expectation maximization, estimation of correlation relationship and dissimilarities, kernel-based Hilbert–Schmidt optimization, optimized features, and classification using the deep learning methodology of RNN—recurrent neural network. Finally, we make the performance analysis with the performance metrics like accuracy, Kappa, TN, TP, FN, FP, precision, recall, Jaccard coefficient, F1-measure, and error.