“…With the established dataset, people have applied a variety of ML algorithms to study MGs with a good glass-forming likelihood [17, 71, 74-76, 78, 79, 85-87] (see figures 3(b), (c) and (e)) or a good GFA [17,[74][75][76][87][88][89][90][91][92] (see figures 3(b), (d) and (f)). These mainly include support vector machine (SVM) [17,74,78,79,92], random forest (RF) [17,66,75,76,90], Gaussian process regression (GPR) [17,74] and artificial neural network (ANN) [17,71,79,87,89]. However, to effectively train an ML model, one has to design proper 'fingerprints' or descriptors for their data (data featurization), as shown in figure 3(b).…”