Predictive quality analytics for the viscosity of water-based architectural paint manufacturing by using improved supervised machine learning and maximum dissimilarity algorithm

Viscosity is a key physical property in the production process of paint. In this research, statistical modeling was used to analyze and predict the viscosity of water-based architectural paint as part of quality control in a coating factory. In this sense, quality technicians constantly seek to improve this indicator. The Viscosity difference is modelled as a function of temperature in the ranges of 19–25 °C and 25–32 °C. Parametric polynomial regression, ANOVA analysis, residual plots, and Box-Cox transformation were used as statistical tools for data analytics and prediction. Model corrections were applied by using Cochrane–Orcutt transformation and assumptions were tested using the Kolmogorov–Smirnov statistics by Lilliefors, Breusch–Pagan, and Durbin–Watson. Improved Maximum Dissimilarity algorithm with the small group filter and representative initial set selection was used for selecting the best representative data to validate the models and three supervised machine learning methods (Random Forest, K-nearest neighbors, and Gradient-boosted trees) were employed through hyperparameter optimization, it was found that Random Forest gave the best performance. Two regression models were obtained: a second-degree polynomial model for samples with a temperature less than 25 °C and a simple linear non-parametric model one for samples at temperature greater than 25 °C. Adjusted coefficients of determination are 0.968 and 0.978, respectively. Finally, using the proposed predictive models could reduce the turnaround time by 48.5%.