Class-Distribution-Aware Unified Prediction Model for Quantification of Quaternary Mixture of Vehicular Exhausts

In this work, class-distribution aware, two-layered unified prediction model has been proposed to classify (layer-1) and subsequently quantify (layer-2) the concentrations of a quaternary mixture of vehicular-exhausts with minimal computational overhead. The proposed scheme utilizes at most one regression model for each gas to learn the signature-response of sensor-array in all gas mixture classes, leading to lesser complexity of the proposed unified model as compared to the earlier reports. Each prediction task has been experimentally investigated on the Google Colab platform using Python programming, revealing that K-Nearest Neighborhood (KNN) multi-class classifier exhibits the highest classification accuracy (F1-score) and minimum computational time, i.e. 96% and 0.39 seconds, respectively. However, the eXtreme Gradient Boosting (XGBoost) and Multilayer Perceptron (MLP) based models quantify NO, C3Hs, N02, and NH3 within the mixture conditions with the maximum coefficients of determination (R2-score), i.e. 0.71, 0.97, 0.95, and 0.94, respectively.