Fig. 2: Prediction quality and interpretation of the presenting imputation-aware, multi-task neural network model.
From: Data-driven material screening of secondary and natural cementitious precursors
a–c Prediction vs. actual values of the three reactivity metrics: heat release, Ca(OH)2 consumption, and bound water. Magenta and gray lines represent the trend of test and train data points, respectively. d–f top ten contributing descriptors (descriptors), with gray bars representing chemical properties, yellow bars representing environmental descriptors, light blue bars representing physical properties, and magenta bars representing mix proportions of additional materials in the paste mix. g–i SHAP values of the top ten contributing material descriptors, where blue and red dots indicate low and high values; R2 represent Coefficient of determination indicating the goodness of fit; top contributing descriptors are identified from the permutation descriptor importance.
