Fig. 3: Construction of a prediction model for stretchable nanocomposites via active learning loops and data augmentation. | Nature Communications

Fig. 3: Construction of a prediction model for stretchable nanocomposites via active learning loops and data augmentation.

From: Predictive design of stretchable electrodes with strain-insensitive performance via robotics- and machine learning-integrated workflow

Fig. 3: Construction of a prediction model for stretchable nanocomposites via active learning loops and data augmentation.

a Schematic illustration of constructing a prediction model via active learning loops, robot–human collaboration, and data augmentation. Created in BioRender. Shrestha, S. (2025) https://BioRender.com/qt9vjwc. b Spatial distribution of 146 data points within the feasible parameter space, including 24 G1–1D, 23 G1–2D, 51 G2–2D1D, and 48 G2–2D2D stretchable nanocomposites. c Using an independent set of testing data points, the MRE values of different prediction models based on LR, DT, GBDT, RF, and ANN committees with 1000-fold training data points by the UIP method. d MRE values of different prediction models based on different virtual-to-real data ratios. e–g Comparison between actual resistance–elongation curves (solid lines) and model-predicted “response” labels (open symbols), including \({\varepsilon }_{5\%}\), \({\varepsilon }_{7.5\%}\), and \({\varepsilon }_{10\%}\). e G1–1D and G1–2D stretchable nanocomposites. f G2–2D1D stretchable nanocomposites. g G2–2D2D stretchable nanocomposites. These 10 stretchable nanocomposites were fabricated under different sets of fabrication parameters (compositions, thicknesses, deformation sequences, and applied pre-strains, as summarized in Supplementary Table 7).

Back to article page