At present, the performance evaluation of lubricating oil is mainly measured by bench tests or on-road tests to measure the friction coefficient and wear rate of the friction pair. However, this method has problems such as long test cycles and high costs. Therefore, in order to experimentally evaluate the lubricating performance of lubricating oil under different degradation conditions and combine it with machine learning technology, a neural network-based lubricating oil performance prediction model is proposed. Firstly, five performance indicators of lubricating oil including water content, viscosity, acid value, Cu, and Fe particle content are selected to design single-factor experiments and multi-factor orthogonal experiments to obtain the friction coefficient and wear rate of lubricating oil under different influencing factors. Then, the backpropagation neural network (BP) and the general regression neural network (GRNN) are proposed, and the experimental data divided into training and prediction sets are input into the neural network for training and prediction. Finally, based on genetic algorithm (GA), the general regression neural network is optimized to obtain a better prediction model. The research results show that compared to the BP prediction model and the GRNN prediction model, the optimized GA-GRNN model has higher prediction accuracy when dealing with small sample, multiple input single output data, with an average prediction accuracy of 97.5% for the friction coefficient and 86.12% for the wear rate.