Prediction of band gap and color of cobalt-doped gahnite ceramic pigment by combining machine learning, DFT, and experimental methods

In recent years, we have conducted some research on the color prediction of ceramic pigments with DFT, and this method was investigated in a review paper. In this work, we selected the gahnite structure and implemented the idea of predicting the color of ceramic pigments and the energy gap using direct and indirect methods with machine learning. Here, we synthesized experimental samples with different doping concentrations. In addition, we modeled theoretical samples, performed DFT+ U calculations, and studied various properties. We generated a dataset of 1066 AB 2 O 4 structures to predict the energy gap and constructed the random forest, gradient tree boosting regression, least absolute shrinkage and selection operator, Ridge, support vector regression (poly, rbf, and linear kernels), and artificial neural network algorithms. After comparing the results, we showed that in some algorithms, the direct method and in others, the indirect method provided more accurate predictions. We calculated the average and dominant (k-means clustering) colors from the digital images of the experimental samples and introduced them as an alternative method to UV–Vis devices. To calculate the color of experimental and theoretical samples, we used code we wrote in Python and examined its performance. Herein, we compared the colorimetric results of the samples. We prepared dataset that included 113 of the CoAl 2 O 4 , , , and Co 2 SiO 4 structures. We used random forest and gradient tree boosting regression methods to predict colorimetric coordinates. We showed that the model we proposed for color prediction was successful.