In molecular design, material design, process design, and process control, it is important not only to construct a model with high predictive ability between explanatory features x and objective features y using a dataset but also to interpret the constructed model. An index of feature importance in x is permutation feature importance (PFI), which can be combined with any regressors and classifiers. However, the PFI becomes unstable when the number of samples is low because it is necessary to divide a dataset into training and validation data when calculating it. Additionally, when there are strongly correlated features in x, the PFI of these features is estimated to be low. Hence, a cross‐validated PFI (CVPFI) method is proposed. CVPFI can be calculated stably, even with a small number of samples, because model construction and feature evaluation are repeated based on cross‐validation. Furthermore, by considering the absolute correlation coefficients between the features, the feature importance can be evaluated appropriately even when there are strongly correlated features in x. Case studies using numerical simulation data and actual compound data showed that the feature importance can be evaluated appropriately using CVPFI compared to PFI. This is possible when the number of samples is low, when linear and nonlinear relationships are mixed between x and y when there are strong correlations between features in x, and when quantised and biased features exist in x. Python codes for CVPFI are available at https://github.com/hkaneko1985/dcekit.
Feature importance (FI) is used to interpret the machine learning model y = f(x) constructed between the explanatory variables or features, x, and the objective variables, y. For a large number of features, interpreting the model in the order of increasing FI is inefficient when there are similarly important features. Therefore, in this study, a method is developed to interpret models by considering the similarities between the features in addition to the FI. The cross-validated permutation feature importance (CVPFI), which can be calculated using any machine learning method and can handle multicollinearity problems, is used as the FI, while the absolute correlation and maximal information coefficients are used as metrics of feature similarity. Machine learning models could be effectively interpreted by considering the features from the Pareto fronts, where CVPFI is large and the feature similarity is small. Analyses of actual molecular and material data sets confirm that the proposed method enables the accurate interpretation of machine learning models.
2 September 2022 in The Journal of Physical Chemistry A
We aim to achieve resource recycling by capturing and using CO2 generated in a chemical production and disposal process. We focused on CO2 conversion to CO by the reverse water gas shift–chemical looping (RWGS-CL) reaction. This reaction proceeds in two steps (H2 + MOx ⇆ H2O + MOx–1; CO2 + MOx–1 ⇆ CO + MOx) via a metal oxide that acts as an oxygen carrier. High CO2 conversion can be achieved owing to a low H2O concentration in the second step, which causes an unwanted back reaction (H2 + CO2 ⇆ CO + H2O). However, the RWGS-CL process is difficult to control because of repeated thermochemical redox cycling, and the CO2 and H2 conversion extents vary depending on the metal oxide composition and experimental conditions. In this study, we developed metal oxides and simultaneously optimized experimental conditions to satisfy target CO2 and H2 conversion extents by using machine learning and Bayesian optimization. We used transfer learning to improve the prediction accuracy of the mathematical models by incorporating a data set and knowledge of oxygen vacancy formation energy. Furthermore, we analyzed the RWGS-CL reaction based on the prediction accuracy of each variable and the feature importance of the random forest regression model.
26 March 2020 in Journal of the American Chemical Society