<p>Soil-rock mixtures (SRMs) are crucial in assessing foundation bearing capacity and slope stability in mountain substation projects, yet their complex mechanical properties pose significant challenges. Current research primarily focuses on the influence of individual factors on shear strength using the controlled variable method, offering limited insight into the interactions and quantitative analysis of multiple influencing factors. This limitation hinders accurate prediction and effective control of shear strength, complicating post-construction settlement management. To address this challenge, this study introduces an analysis method employing a feedforward neural network (FNN) to evaluate changes in SRM shear strength under the influence of multiple factors. By integrating SRM physical properties with shear strength test data, the method facilitates identification and quantitative analysis of key factors, including moisture content, dry density, void ratio, and the liquid-plastic limits of fillers with varying particle gradations. A correlation model is developed, demonstrating high reliability and predictive accuracy. Among the analyzed factors, moisture content and plastic limit exhibit the most significant influence on SRM shear strength, with importance levels ranging from 23.9% to 32.8%. The primary contribution of this research is the integration of machine learning with traditional geotechnical analysis, offering a practical framework for the identification and evaluation of factors influencing SRM shear strength. The proposed correlation model provides valuable insights into the intrinsic factors governing SRM shear strength variability and offers practical guidance for the design, construction, and operational safety of fill engineering in substations across southwestern China. Moreover, it serves as a useful reference for the quantitative analysis of influencing factors in SRMs for other regions.</p>