Fig. 4: Results and insights from ML model. | Nature Communications

Fig. 4: Results and insights from ML model.

From: Universal machine learning aided synthesis approach of two-dimensional perovskites in a typical laboratory

Fig. 4

a Schematic sketch of the problem-specific descriptors. Here, \({d}_{{{{{{{\rm{N}}}}}}}_{{{{{{\rm{i}}}}}}}-{{{{{{\rm{Atom}}}}}}}_{{{{{{\rm{j}}}}}}}}\) represents the topological distance between the nitrogen i and atom j in the molecular skeleton, \({d}_{{{{{{{\rm{N}}}}}}}_{{{{{{\rm{i}}}}}}}-{{{{{{\rm{N}}}}}}}_{{{{{{\rm{j}}}}}}}}\) represents the topological distance between nitrogen i and j. b Receiver operating characteristic (ROC) curve and confusion matrix for the synthesis feasibility of 2D perovskites. c The sorted mean SHapley Additive exPlanations (SHAP) values of selected features in the ML model. d SHAP values for six features of the ML models, plots with different colors represent different features. e SHAP analysis of (ClC6H4CH4NH3)4AgBiI8, (BrC5H4NH)4AgBiI8, and (NH2C4H7NHC2H6)2AgBiI8. Red and blue arrows represent the positive and negative contribution of features on the synthesis feasibility of 2D AgBi iodide perovskites, respectively. The expected base value of synthesis feasibility is −0.0138, and the ML-predicted synthesis feasibility of each sample is bolded in black.

Back to article page