Abstract: For the accurate identification of liquid loading severity in gas wellbores and the optimization of drainage gas recovery strategies, this study addresses the lack of quantitative classification criteria in field applications. By employing the K-means clustering algorithm, an improved silhouette coefficient method is proposed to determine the optimal number of clusters K with higher precision. Using production data from 86 wells in a certain oilfield, a dual classification criterion integrating tubing-casing pressure difference and daily gas production was established to categorize liquid loading levels. The results show that improved silhouette coefficient method achieved an accuracy rate of 94.6% in selecting the optimal K value. During the classification process, the severity of liquid loading is firstly divided into four main categories based on the tubing-casing pressure difference Δp: zero liquid loading (Δp＜2 MPa), mild liquid loading (2 MPa≤Δp＜3.41 MPa), moderate liquid loading (3.41 MPa≤Δp＜5 MPa), and severe liquid loading (Δp≥5 MPa). And then, for the three categories with liquid loading, the subdivision is made on the basis of daily gas production: mild liquid loading corresponds to daily gas production ranges of (1.15~1.38) × 10⁴ m³/d and (0.47~1.15) × 10⁴ m³/d, moderate liquid loading corresponds to (0.38~0.47) ×10⁴ m³/d and (0.35~0.38) ×10⁴ m³/d, and severe liquid loading corresponds to (0.26~0.35)×10⁴ m³/d and (0~0.26)×10⁴ m³/d. After applying this dual criterion, the accuracy of liquid loading severity identification has increased to 88.4% from 62.8%. This quantitative classification standard can effectively support the selection of drainage gas recovery technologies and parameters optimization, and plays guiding role in improving the effect of measures and the efficiency of fluid drainage.