Machine Learning-Assisted Estimation of the Photoantioxidant Activities of Bare, Mg, Cu, and Mg/Cu Dual-Doped ZnO

Green-synthesized pure zinc oxide (ZnO), Mg-doped ZnO,
Cu-doped
ZnO, and Mg/Cu dual-doped ZnO using aqueous leaf extract of Ziziphus mauritiana were analyzed for their antioxidant
activities. In this study, the data-driven approach has been used
to estimate the photoantioxidant activities of ZnO, Mg-doped ZnO,
Cu-doped ZnO, and Mg/Cu dual-doped ZnO based on the experimental data
and synthetic data generated through simulations. Three different
machine learning models, including artificial neural network, extreme
gradient boosting, and automated machine learning, were explored and
compared for both data sets. These models were validated by using
external validation and applicability domain methods based on the
values of coefficient of determination, root mean square, and mean
absolute errors. The performance of the machine learning techniques
showed that photoantioxidant activities could be predicted accurately
from the input variables such as types of dopants, percentage of dopants,
average crystallite size, lighting condition, and concentration of
antioxidants (photocatalyst). Doping and the lighting condition were
found to have a more significant impact on the values of photoantioxidant
activities of the ZnO, Mg-doped ZnO, Cu-doped ZnO, and Mg/Cu dual-doped
ZnO in comparison to other variables. Based on three artificial neural
network models, the variables for Mg doping and the lighting condition
had weights with values ranging between 1.1 and 2.9.