Prediction of affinity coefficient for estimation of VOC adsorption on activated carbon using V-matrix regression method

Volatile organic compounds (VOCs) pose an ever-growing threat on human health and environment. Predicting VOC affinity coefficient and consequently estimating activated carbon adsorption requires fundamental understanding of adsorbate-adsorbent interaction, shape and hydrophilicity effects. Hence, a model which expressed these three factors with molecular descriptors was investigated with Ridge and V-matrix regression methods in two cases, k-fold cross-validation and random sampling technique. The results showed, the sole interaction term and the complete model decreased the root mean square error (RMSE) of polarizability ratio by approximately 19% and 26% respectively. The V-matrix regression, reduced the average Ridge RMSE by 9 and 19% for the first and second case. Lower than 10% errors were displayed by 104 out of 155 data and only 4 data which were small molecules with very high polarity had more than 30% error. For both cases the proposed model with V-matrix regression had better or similar results compared to previous research. However, the effect of reference compounds on highly polar VOCs requires further investigation. From the VOC adsorption estimation, it was evident that affinity and adsorption errors were in the same magnitude. Hence, with accurate prediction of affinity coefficient, adsorption isotherms of any VOC can be calculated.