Hybrid model with optimal features for non-invasive blood glucose monitoring from breath biomarkers

Abstract

Analysis of exhaled breath is becoming more and more used as an additional diagnostic technique in medicine. Researchers must create unique algorithms for accurate data interpretation due to the sheer quantity of factors that must be considered. Therefore, a new NICBGM-based model from exhaled breath is introduced in this study. This work exploited median filtering (MF) for pre-processing. Then, “Improved Empirical Wavelet Functions (I-EWF), R-peak detection, QT intervals, PR intervals, Entropy-based feature, improved Discrete wavelet transform (DWT), Continuous Wavelet Transform (CWT), and short-time Fourier transformation (I-STFT)” are extracted. Further, optimal features are chosen, which are then put through a hybrid scheme that combines “Deep Max out (DMO) and Long Short-Term Memory (LSTM)”. Then, the mean is taken by DMO and LSTM to attain the fine result. Here, the Wild Beest Updated HGSO (WBU-HGSO) model is used to optimize the LSTM weights. The final step is an analysis that proves the superiority of the WBU-HGSO-based model.