Abstract: To address the problem of excessive deformation in roadway surrounding rock caused by the high-stress geological environment of deep coal mines, and to overcome the core contradiction in traditional support—being “rigid but prone to failure, yet flexible and difficult to control”—this study aims to achieve precise matching of support parameters and stable control of roadways. Taking the Yuecheng and Zhaozhuang coal mines as the engineering background, the research investigates the control effect of synergistic anchoring support and an intelligent decision-making model for precise support parameters by combining field investigation, theoretical analysis, FLAC^3D numerical simulation, and machine learning. A numerical simulation database comprising 24 working conditions is constructed. By comparing three algorithms—BP Neural Network（BPNN）, Support Vector Regression（SVR）, and Random Forest（RF）—an intelligent decision-making model is optimized and constructed, a “locally targeted” optimization strategy is proposed, and industrial trials are conducted. The results show that the traditional single support has obvious limitations. The peak stress of full-length anchoring is as high as 693.5 kN, exceeding the 650 kN fracture threshold, and brittle fracture is very likely to occur due to stress concentration. Although extended anchoring does not fracture, the roof and floor deformation reaches 1 199.4 mm, resulting in continuous rheology of the surrounding rock. Synergistic anchoring achieves a combination of rigidity and flexibility, effectively controlling the maximum deformations of the two ribs and the roof-floor at 174.5 mm and 845.2 mm respectively, with a significant reduction of 25% to 48%. In addition, the cable stress is stably controlled at about 600 kN, retaining a safety margin of approximately 50 kN. The RF algorithm demonstrates a significant advantage with an R² exceeding 0.95 and a MAPE of only 0.26%, far superior to the 3.24% of BPNN and 2.93% of SVR. The model accurately identifies four breakage risks with a peak of 701.3 kN. In industrial trials, the optimized scheme reduces the maximum stress of the anchor cable to 592.8 kN, which remains consistently below the 650 kN breaking threshold, ensuring the safe extraction of the working face. This research provides an economically efficient and differentiated support solution for deep high-stress roadways and promotes the transformation of support design toward a data-driven paradigm.