Environmental Advances (Oct 2022)
Novel Machine Learning (ML) models for predicting the performance of multi-metal binding green adsorbent for the removal of Cd (II), Cu (II), Pb (II) and Zn (II) ions
Abstract
The present work embodies the implementation of Machine Learning (ML) techniques to predict the adsorption breakthrough curves of four heavy metals (Cd, Cu, Pb and Zn) over low cost multi-metal binding bioadsorbent (MMBB). The MMBB is a 3:2:1 dry weight ratio of tea waste, maple leaves, and mandarin peel. Two ML techniques namely, Support Vector Regression (SVR) and Artificial Neural Networks (ANN) are employed for model development. Also, the conventionally employed multiple linear regression (MLR) technique is used with the same data set. The relative concentration (CtCi) as a function of input independent variables namely, bed depth, initial concentration, flow rate and time is predicted by the said models for each metal ion. Statistical evaluation parameters used for model evaluation are coefficient of determination (R2), root mean square error (RMSE), mean relative error (MRE) and average absolute relative error (AARE). The R2 values for developed cadmium, copper, lead and zinc MLR models are 0.831, 0.8676, 0.8739, and 0.8567, respectively whereas its corresponding values obtained for SVR models are 0.9981, 0.997, 0.998, and 0.997, respectively. The values of the parameter AARE calculated for the SVR models for the four different metals (Cd, Cu, Pb, and Zn) are 0.0586, 0.102, 0.063, and 0.176, respectively while the values of this parameter for the ANN models developed for each of the aforementioned metals in that order are 0.0901, 0.0911, 0.0683, and 0.2069 respectively. Among the ML models, both SVR and ANN models are found to be highly accurate and generalized. On varying the input parameters, the breakthrough curves predicted by the SVR and ANN models lie in close proximity to the experimental curves, whereas those predicted by the MLR models show a large deviation from the experimental values. Thus, ML models can be fruitfully applied for predicting the breakthrough curves of heavy metals.
