Modeling methylene blue removal using magnetic chitosan carboxymethyl cellulose multiwalled carbon nanotube composite with genetic algorithms and regression techniques.

The purpose of this study was to model and optimize the removal of methylene blue using a novel magnetic chitosan-carboxymethyl cellulose/multiwalled carbon nanotubes and to identify the most significant parameters influencing the adsorption efficiency. Genetic Algorithm and other statistically advanced techniques such as Gradient Boosting Regressor, and Maximum Likelihood Estimation were used to extract the necessary process parameters and the factors that posed a major impact on the adsorption efficiency. The following metrics showed the secondary model, trained using the Gradient Boosting Regressor technique, had a slightly better accuracy of 0.99, Root Mean Square Error of 0.68 and Mean Absolute Error of 0.49 compared to the Maximum Likelihood Estimation of 0.94 in the training sample and 0.95 in testing. Gradient Boosting Regressor model was more stable and did not overfit, there was some sign of overfitting in Maximum Likelihood Estimation. From the feature importance X2 (initial methylene blue concentration) was the most important feature while X1 (contact time) and X4 (adsorbent amount) were not too important and can be eliminated from the models. The result of Genetic Algorithm analysis also proved the model has converged to the optimal solution effectively, the best solution of X1 = 49.41, X2 = 110.62, X3 = 11.85 and X4 = 20 which gives the maximum removal efficiency = 94.64% of methylene blue. These steps also supported the increased importance of X2, with a positive coefficient of 0.72 for improved removal efficiency as well as X3 which correlated positively with a coefficient of 0.66 in this regard. The adsorbent showed stability in residuals in the training set equal to Mean Residual = 0, and Root Mean Square Error of 0.68, while testing gave the Mean residual = 0.15, and the Root Mean Square Error of 2.33. The major conclusion drawn is that the algorithm working on the Gradient Boosting Regressor was more efficient, and had a higher accuracy margin as well as a more stable model than Maximum Likelihood Estimation. The result indicated that the adsorbent possessed a higher removal efficiency of 94.64%, thus, its application in the removal of dyes from wastewater could be seen as possible.