To address the reliance of shearer gearbox oil maintenance on manual experience and the difficulty of quantitatively evaluating multi-feature data, a multi-feature data-driven oil maintenance decision system is developed. The system integrates oil temperature, physicochemical indicators, and ferrography detection data, extracting 14 characteristic parameters, and proposes three key technologies: (1) a wear trend recognition method based on recurrent neural networks (RNN), which employs a sliding window to capture temporal dependencies; (2) a wear stage classification method based on bidirectional gated recurrent units (Bi-GRU), which learns bidirectional temporal relationships of features; and (3) fault analysis and recommendation by combining a three-level quantitative scale with hierarchical logical reasoning, establishing associations between degradation parameters and maintenance suggestions. Experimental results show that the RNN-based wear trend recognition achieves an accuracy of 90.7%, and the Bi-GRU-based wear stage classification reaches 91.49%; field validation demonstrates consistency with expert opinions. Furthermore, test bench experiments confirm the system’s generalization capability. The proposed system effectively integrates multi-source heterogeneous data, enabling intelligent recognition of shearer gearbox wear states, fault analysis, and maintenance recommendations, thereby significantly reducing reliance on manual experience and improving the objectivity and reliability of maintenance decision-making.