Fig. 3: Feature engineering and ML model training based on the VIIInfo dataset. | npj Computational Materials

Fig. 3: Feature engineering and ML model training based on the VIIInfo dataset.

From: Exploring high-performance viscosity index improver polymers via high-throughput molecular dynamics and explainable AI

Fig. 3

a Process of descriptor down-selection, The thresholds for variance, Pearson, Spearman, Distance correlation coefficients and the maximum information coefficient (MIC) were 0.1, 0.05, 0.05, 0.105, and 0.195, respectively; b Pearson correlation coefficient heatmap matrix between optimized descriptors and target viscosity property; c R2 scores from 30 independent training runs of four different ML algorithms based on optimized descriptors; d Accuracy assessment of the XGBoost model trained on optimized descriptors.

Back to article page