Evolving an Accurate Decision Tree‐Based Model for Predicting Carbon Dioxide Solubility in Polymers

Abstract

Solubility is one of the most indispensable physicochemical properties determining the compatibility of components of a blending system. Research has been focused on the solubility of carbon dioxide in polymers as a significant application of green chemistry. To replace costly and time‐consuming experiments, a novel solubility prediction model based on a decision tree, called the stochastic gradient boosting algorithm, was proposed to predict CO2 solubility in 13 different polymers, based on 515 published experimental data lines. The results indicate that the proposed ensemble model is an effective method for predicting the CO2 solubility in various polymers, with highly satisfactory performance and high efficiency. It produces more accurate outputs than other methods such as machine learning schemes and an equation of state approach.

Augmenting Polymer Datasets by Iterative Rearrangement

“…The CO 2 solubility dataset has 515 data points, 13 unique polymers, and three input features: molecular representation, pressure (mPa), and temperature (K) in a sorption cell for a pressure-decay experiment. , The target property is CO 2 solubility in polymers (g of CO 2 dissolved/g of polymer).…”

Augmenting Polymer Datasets by Iterative Rearrangement

“…The CO2 Solubility dataset contains 13 different polymers common to gas solubility experiments (Table S1). 34 The two continuous variables, temperature and pressure, have a wide distribution of values (Figure S1), which is suitable for ML because the data is sufficiently diverse for the models to extract information from it. Because temperature and pressure are strongly correlated with the solubility of CO2 independent of polymer structure (Figure S2), all model-representation pairs are relatively accurate (most R 2 > 0.7).…”

“…Soleimani et al trained a new model for each polymer using its corresponding data. 34 Therefore, the resulting models cannot generalize beyond the specific polymer used for training. For the sake of comparison, we calculate a global R 2 value across all the model predictions for each polymer.…”

“…The CO2 Solubility dataset has 515 data points, 13 unique polymers, and 3 input features: molecular representation, pressure (mPa), and temperature (K) in a sorption cell for a pressure-decay experiment. 34,53 The target property is CO2 solubility in polymers (g of CO2 dissolved / g of polymer).…”

“…First, we compare the performance of our augmented representations to that of common molecular representations with well-established ML and deep learning (DL) models across three different datasets. These datasets cover different structureproperty relationships in polymers: CO2 Solubility, 34 Pervaporation Separation, 35 and Swelling. 36 To this end, we built a rigorous ML workflow which incorporates stratified cross-validation, and automated hyperparameter optimization.…”