Machine learning approach to predict the viscosity of perfluoropolyether oils

Perfluoropolyethers (PFPEs) have attracted much attention due to their exceptional chemical stability, thermal resistance, and wide application in high-performance industries such as aerospace, semiconductors, and automotive engineering. One of the most important properties of PFPEs as lubricants is their viscosity. However, experimental determination of viscosity is time-consuming and expensive. In this study, four intelligent models, Multilayer Perceptron (MLP), Support Vector Regression (SVR), Gaussian Process Regression (GPR), and Adaptive Boost Support Vector Regression (AdaBoost-SVR), were used to predict the viscosity of perfluoropolyethers based on parameters such as temperature, density, and average polymer chain length. Statistical error analysis showed that the GPR model had higher accuracy than other models, achieving a root mean square error (RMSE) of 0.4535 and a coefficient of determination (R 2 ) of 0.999. To evaluate the innovation and effectiveness, we compared the GPR model with the predictions obtained from the traditional Waterton and Vogel-Fulcher-Thamman (VFT) correlations. Furthermore, the GPR model not only provided high accuracy, but also successfully captured the underlying physicochemical trends. To further evaluate the performance of the GPR model, the leverage technique was utilized, which demonstrated that 98.33% of the data points fell within the valid range, emphasizing the robustness and reliability of the model.