Figure 2
From: Artificial intelligence inferred microstructural properties from voltage–capacity curves
Aggregated true and predicted scatter plots from tenfold cross validation for (a) Bruggeman’s exponent \(\alpha\) from \(L_{\mathrm {w}}\)-model with \(L_{\mathrm {s}} = 5.59 \%\) and \(R^2 = 0.99\), (b) \(\alpha\) from \(L_{\mathrm {M}}\)-model with \(L_{\mathrm {s}} = 2.95 \%\) and \(R^2 = 0.99\), (c) shape factor S from \(L_{\mathrm {w}}\)-model with \(L_{\mathrm {s}} = 0.89 \%\) and \(R^2 = 1.0\), and (d) S from \(L_{\mathrm {M}}\)-model with \(L_{\mathrm {s}} = 2.59 \%\) and \(R^2 = 0.97\). Overall, the trained model accurately predicts both \(\alpha\) and S. Specifically, the \(L_{\mathrm {w}}\)-model performed better at predicting S with 1.70% less error throughout the range of S values, while the \(L_{\mathrm {M}}\)-model was better at predicting \(\alpha <3.0\) by over 5% in comparison to the \(L_{\mathrm {w}}\)-model. A combination of the model predictions was adopted for the final prediction.
