Machine learning-based heat deflection temperature prediction and effect analysis in polypropylene composites using catboost and shapley additive explanations

Abstract

Among the various physical properties of polypropylene composites (PPCs), heat deflection temperature (HDT) during PPC production is significant because it is directly related to the mechanical behavior of such products. However, it is difficult to predict and analyze the HDT of PPCs owing to the absence of a mathematical or theoretical model. Moreover, categorical data which has different physical properties despite using same substances made predictions highly difficult in PPCs. Here, this study proposed an indicator to analyze the categorical data and Catboost-based model for HDT prediction considering the categorical data. First, the categorization and minimum-based values (MBVs), a dimensionless factor used to calculate HDT differences in the categorical dataset, are applied to detect and split categorical data in a PPC dataset. Second, a case study was conducted on the dataset using three algorithms to compare the proposed model with other traditional machine-learning approaches. As a result, the proposed model provides the highest prediction performance of R2=0.8965 and 0.9801, for the total test dataset and the categorical dataset, respectively. In addition, the effect analysis of substances on HDT was conducted to get some prospective substances for the required HDT using Shapley Additive Explanations (SHAP). Thus, the results of this study can provide a guidance for selecting prospective substances using SHAP result for the target HDT and adjusting the substance ratio using the proposed model. It is expected that this framework has the potential for being applied to the other blending processes to produce products with the required properties.