Network Equivalent of Voltammetric Sensors as Quality Descriptors of Tea Samples With Leave-P-Out Cross-Validation

This work presents an extended approach towards quality assessment of liquid analytes using equivalent electrical circuit parameters obtained through an established method of system identification of voltammetric sensors (E-Tongue). To demonstrate the application of this work, the equivalent electrical parameters (resistances and capacitances) are extracted from the electrochemical (electrode-electrolyte) system considering 20 independent black tea samples as analyte. These parameters are then correlated with the concentration of a well-known quality descriptor; theaflavin compound in tea, thereby facilitating a quantitative and non-destructive method for evaluating tea quality using E-Tongue. In this direction, calibration models based on linear as well as nonlinear regression techniques have been developed. The regressors have been trained with the circuit parameters as features to predict the theaflavin content in tea samples. These prediction results are compared with data models that used existing feature extraction (selection) methods, such as discrete wavelet transform (DWT) and neighborhood component analysis (NCA). Additionally, leave-P-out cross-validation (LPOCV) has been implemented to obtain theaflavin prediction models for different combinations of train-test datasets, ensuring a robust and comprehensive assessment of tea quality. For a given train-test combination of data set, the LPOCV achieved a prediction accuracy as high as 94.80% using the proposed feature extraction method, outperforming both DWT and NCA. Statistical analyses revealed that the equivalent circuit parameters effectively map the tea quality, even when evaluated across various combinations of data sets. The approach offers a practical solution for monitoring tea quality at a sample level, potentially reducing reliance on subjective sensory methods.