Applicability of Regression Technique for Physical Modeling: a Case Study on Adsorption in Wastewater Treatment

The reliability of Physical Modeling in applications such as Adsorption and Heat transfer studies is not accurate since their mechanisms are complex and a proper understanding of the physics of the system is incomplete. In order to verify the applicability of Regression technique for Physical Modeling, a physical model is developed based on Multiple regression technique to predict the Pollutant Removal efficiency of fluoride in adsorption studies. Two sets of data points are collected viz., of twenty-one points consisting of homogeneous data with respect to adsorbent and of forty-eight points (heterogeneous data, including the above twenty-one points) and tested with the model. Results showed that, the physical model is giving encouraging results for homogeneous data (Standard Deviation (SD): 0.157) but is giving erratic results (SD: 0.361) for the heterogeneous data. The heterogeneous data consists of non-linear adsorption data, which the model could not predict accurately indicating that, the Regression technique holds a limitation in understanding the physics of the system. Novel techniques such as ANN can be used to predict the output from the data set with better accuracy than that using Regression technique. Back propagation Network of ANN is used as a test trial for the above database and the results are encouraging (SD: 0.29) with respect to heterogeneous data.