Base oil is the core of almost any lubricant and is one of the specialty products that can be produced in an oil refinery. Base oil processing is a complex process consisting of multiple units such as the atmospheric distillation unit, the vacuum distillation unit, intermediate tanks, the treating unit, the isomerization unit, the dearomatization unit, and the product fractionation unit. Variations in the properties of the different crude oil feedstocks, combined with the complexity of the integrated processing units with batch/continuous processes, result in a significant challenge in maximizing the base oil production to meet increasing consumer demands. Typical solutions to improve the yield of the base oil adopted in practice are to manually adjust the operating conditions based on past experiences. However, manual adjustment is not trivial, as changing one parameter affects several base oil properties. Thus, only one operating condition is changed at a time, while others remain fixed, resulting in a substantial delay in assessing the impact of the changes made. Consequently, production loss, product quality deterioration, waste generation, and higher energy consumption are inevitable. In addition, most planning analyses reported in the literature tend to assume a fixed yield of the base oil for each grade, regardless of feedstock quality. However, the yield function of the base oil is greatly dependent on the feed properties and operating conditions. Hence, in this paper, an attempt is made to automate the base oil yield maximization of an integrated industrial base oil processing complex via a machine learning-based optimization framework. Five machine learning models are explored: decision tree regression, support vector regression, artificial neural networks, random forest regression, and extreme gradient boosting (XGBoost) to represent the highly integrated industrial base oil processing complex. The XGBOOST models were found to be superior in performance and subsequently introduced as part of optimization functions to the global Bayesian optimization and differential evolution optimization techniques. The resulting machine learning-based optimization framework has been successfully validated during the actual plant trial at the integrated industrial base oil complex from January 2020 to December 2021, and the results show considerable improvement in yield further by 5.24 and 4.48 for both viscosity index optimization cases using Bayesian optimization and differential evolution techniques, respectively, and 4.02 and 2.05 for pour point optimization cases using Bayesian optimization and differential evolution techniques, respectively. © 2023 American Chemical Society.