In this article, the authors have experimentally investigated the changes in four key properties of six non-edible low-impact energy carries based on rapeseed oil quality grade HM and viscosity grade VG46, which were used as a filling in the hydraulic system of a round wood sorting and transporting trolley. These oils were enriched with thermo-oxidizing, extreme-pressure additives, anti-foaming, and lubricating additives to enhance performance. Three supervised machine learning prediction algorithms were used to predict key parameters essential for optimizing their performance and RUL (remaining useful life), namely support vector regression (SVR), generalized additive model (GAM), and Gaussian process regression (GPR). The model’s performance was scored from multiple perspectives using metrics such as root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and coefficient of determination (R2) to state actual values, thereby demonstrating the validity of the models in predicting lubricant lifespan. Based on the collected data, this study demonstrated that it is possible to predict the degradation of hydraulic oil factors to the limit state, integrate these parameters into a comprehensive metric for more accurate remaining useful life (RUL) estimation, and obtain actual operating trends. A negative correlation was found between the remaining useful life (RUL) and parameters such as acid number, kinematic viscosity, peroxide number, and water content. The comparison of modeling algorithms showed that all three algorithms adequately described the degradation patterns. By using these performance criteria, we defined the most accurate and reliable soft-computing model for predicting hydraulic fluid parameters, providing valuable insights into optimizing machine learning models for practical applications.