English abstract
The environmental pollution caused by antibiotics such as amoxicillin and ibuprofen has
gained increasing attention in recent years. These compounds should be eliminated from
discharged effluents to prevent possible negative effects on humans, animals, and the
environment. Recently, adsorption processes have gained more and more interest among
the possible methods to remove emerging pollutants from water. In this study, activated
carbon and potassium persulfate has been investigated for the adsorption of amoxicillin
and ibuprofen from synthetic wastewater. Scanning Electron Microscopy (SEM), Fourier
transform infrared (FTIR) and X-ray diffraction (XRD) analysis has been used in
adsorbents characterization. Furthermore, the adsorption mechanisms, performance,
kinetics and isotherm for amoxicillin and ibuprofen removal were evaluated. To optimize
the experiment results by minimizing the final concentration and increasing the percentage
of amoxicillin and ibuprofen removal efficiency, Response Surface Methodology-Central
Composite Design (RSM-CCD) (statistical analysis) and Support Vector RegressionGenetic Algorithm models (SVR-GA) (machine learning) were used by Minitab and
MATLAB software.
The removal process was conducted using different pH values (3–9), adsorbent dosages of
activated carbon and potassium persulfate (10–50 mg), contact time (5–120 min) and initial
concentration of each antibiotic (5-200 mg/L). It was found that both adsorbents removed
amoxicillin and ibuprofen effectively at pH 7 and the optimal contact time was 5 minutes.
It has been demonstrated that the Langmuir isotherm model for both AMX and IBU
removal accurately defines the isotherm data (R2 = 1). In addition, a pseudo-second-order
model was used to describe the adsorption kinetics. As a consequence of these data, the
maximum adsorption capacity of amoxicillin on activated carbon and potassium persulfate
was 90.97 and 196.31 mg/g, respectively, while for ibuprofen it was 135.23 and 196.47
mg/g.
In the modelling part, RSM and SVR-RBF models could be fitted to both models to predict
the results of the final concentrations of amoxicillin and ibuprofen. The performance of
these models is evaluated using mean absolute relative error (MARE%). Both prediction
models have identical MARE equal to or less than 25%. In this case, the final concentration
can be predicted using either the GA-SVR or the RSM model. On the other hand, the value
of R2
obtained by RSM and SVR was close to 1, but the lowest mean absolute relative
error (%) for all antibiotics and adsorbents is modelled by machine learning, which is better
than RSM-CCD.
