Figure 3
Classifier development procedure. (a) Multi-label data transformation to the binary relevance format. The first point group (\(\mathrm {Y}_1\)), is taken and fed to the binary classifier algorithm. The next binary data \(\mathrm {Y}_2\) is processed using the same procedure till the last binary data \(\mathrm {Y}_{32}\). If the point group cell is 0, the chemical formula at that row does not exist with this symmetry and vice versa. The configuration of each point group 0 or 1 varies among the considered material space for training; and (b) The implemented 5-fold cross validation approach for each binary classification problem. The testing data set varies from one fold to another. After that, the training data set is resampled and subsequently trained for each fold. Then, the training model is quantified using the unseen testing data set. Finally, the evaluation metrics of the five folds are averaged to obtain the final result.
