Phononic crystal-based pH sensing and its classification with machine learning

초록

We present a 3D phononic crystal-based pH sensor made of a biocompatible hydrogel-SiO2 composite, its classification via machine learning (ML), and theoretical and experimental evidence for the existence of a band gap (BG). Instead of spatial, spectral approach is considered to classify different pH levels. The transition in band diagram and transmission coefficient measured along [111] direction reveals that the composite is sensitive to pH variation. Dips showing the BG in transmission spectra shift towards lower frequencies as the pH value increases. In addition, we have demonstrated the potential of four ML algorithms: k-nearest neighbor (KNN), Na & iuml;ve Bayes, Support vector machine (SVM) and Neural Networks (NNs) as a quick identifying tool for the expedited and accurate classification of pH values with sensitivity S=0.3508 MHz/pH up to resolution of R= 0.1 pH level change with only 53 observations. For boosting the predicting power of model, we apply data augmentation technique on dataset. Our work focuses on multiclass supervised ML techniques which have the ability to classify pH values between 4 and 7 in 31 labels. In this comparative study we have explored the capability of each algorithm based upon the performance metrics. In addition, we found that except Na & iuml;ve Bayes all algorithms could predict with >99 % accuracy. Lastly, globally optimal hyperparameters for NNs are also presented. The proposed pH sensor and classification method has pronounced potential to be used for in-vivo powerless pH measurements.

키워드