Introducing a framework for modeling of drug electrochemical removal from wastewater based on data mining algorithms, scatter interpolation method, and multi criteria decision analysis (DID)

Abstract

The contamination of waters by persistent organic pollutants, especially pharmaceutical contaminants, is one of the concerns all over the world. To date, among the treatment methods, the efficient EAOPs method have shown a high ability to treat this type of pollutant. However, conducting adequate tests for taking into account almost all possible conditions to predict the amount of pollutant removal in different conditions is still a challenge. On the other hand, achieving this aim requires a lot of cost and time. The superiority of data mining based methods over conventional mathematical methods have made these methods a good solution to solve this problem. Hence, in present study a model by employing data mining algorithms includes Artificial neural network (ANN), adaptive neuro-fuzzy inference system (ANFIS), M5 model tree, least-square support vector machine (LSSVM) and hybrid of LSSVM and firefly optimization algorithm (FFA), scatter interpolation method, and multi-criteria decision making, namely DID, is presented for modeling of drugs removal. For this purpose, four different inputs include current density, electrolyte concentration, pH, and electrolysis time are used for electrochemical removal of Ciprofloxacin (CIP) as a model pollutant. Subsequently, Scatter interpolation method is used for generating enough data for more accurate modeling and more reliable results. In the final part of the survey, the TOPSIS method under six scenario, is employed for ranking of algorithms by considering accuracy and time criteria. In defined scenarios for TOPSIS, six different weights are considered for time criteria as well as the weights of accuracy are considered as equal in each scenario. Also, the sum of scores of each algorithm in all scenarios is used for final decision. The finding results by TOPSIS for original data showed the superiority of LSSVM_FFA. After generating new data, the M5 and the ANFIS have better results in 0.25 time weight. However, by decreasing time weight and increasing accuracy weight (after second scenario), the M5 and the LSSVM_FFA have better results. Besides, based on the sum of scores for new data, the M5 and the LSSVM_FFA have superiority. Finally, it can be concluded that M5 in about to 3 s, and LSSVM_FFA in about to 17 s lead to estimate the drug removal value with good accuracy, and without needing to high cost, and several months laboratory works. Therefore, the mentioned models can be used for different tasks, such as determining the optimal removal of drug, and investigating the impact of different parameters on drug removal process, without needing to each special experiment. Thus, for generating the large data set, the results of the present study are reliable.