Abstract: Current evaluation methods for single-borehole gas extraction status typically rely on gas concentration, while overlooking the diversity of coal seam gas occurrence. Supervised learning models depend on labeled sample features, but manual labeling becomes costly when the sample size is large. Unsupervised learning models lack sample labeling, making qualitative evaluation infeasible. To address these issues, an evaluation method based on semi-supervised learning was proposed for the gas pre-extraction status evaluation of coal seam boreholes. A multi-dimensional evaluation system was established, incorporating eight indicators such as methane concentration, extraction negative pressure, and ambient temperature. The weighting method combining the analytic hierarchy process (AHP) and fuzzy evaluation method (FEM) was used to establish classification standards for extraction performance. Building on this, a semi-supervised learning model based on the Gaussian mixture model (GMM) and K-Means algorithm (SSGMM/SSK-Means) was developed. By integrating a small number of manually labeled samples and a large quantity of unlabeled data, the model enabled dynamic classification of single-borehole extraction status. The SSGMM demonstrated better clustering rate, while the SSK-Means achieved higher efficiency, developing a complementary "accuracy-efficiency" relationship. The application results from the 215 working face of the Huangling No. 2 Coal Mine in Shaanxi Province showed that the maximum validity clustering rate (MVCR) and adjusted rand index (ARI) of SSGMM and SSK-Means reached 82.64% and 85.83%, respectively, significantly outperforming conventional clustering methods. After optimization through a dynamic feedback mechanism, boreholes initially classified as "poor" showed an improvement of 5.26% to 5.80% in extraction efficiency, achieving a 100% remediation rate.