Notes. The performance is given in each cell position as the bias followed by the dispersion (defined as in Fig. 2): (b,  σ) [dex]. We note the similar performance between the ANN and RF methods. Moreover, we note the significant improvement when moving from central to group parameters (especially for multi-parameter groups, as seen in Fig. 2), but the lack of additional improvement moving to the “all” parameter set (as seen in Fig. 1). This serves to justify our choice in parameters and machine learning method.