Fig. 6: Parity plots and error distributions for three regression models. | npj Computational Materials

Fig. 6: Parity plots and error distributions for three regression models.

From: A database of experimentally measured lithium solid electrolyte conductivities evaluated with machine learning

Fig. 6

AutoSklearn models assessed under LOCO-CV (a), (d) share the most similarity to the controls in Fig. 5, and are thus judged to be the least effective ML model under investigation. Under LOCO-CV, CrabNet models with transfer learning offer improved performance, which can be visually confirmed by the spread of points falling closer to the leading diagonal (b), and the distribution of errors being centred around 0 with a smaller standard deviation (e). ML models give a tighter distribution of errors when validated with k-folds, with transfer learned CrabNets possessing the most favourable actual vs. true characteristics (c) and distribution of errors (f).

Back to article page