Abstract: In order to improve landslide susceptibility assessment for gas transmission pipelines in the Sichuan–Chongqing region, a hybrid transfer learning framework was developed to investigate pipeline-related landslide risks in data-scarce mountainous areas. Verified landslide records from pipeline patrols in the Southwest Oil & Gas Field were used as source domain data, while key environmental variables—such as slope aspect, land use, slope gradient, NDVI, lithology, distance to faults, and seismic acceleration—served as input features. Four machine learning algorithms, including Support Vector Machine (SVM), Isolation Forest (IF), Local Outlier Factor (LOF), and Gaussian Mixture Model (GMM), were integrated through transfer learning to enhance prediction generalization across different regions. The results show that pipeline landslides are mainly associated with southeast-facing slopes, moderate gradients (~10°), NDVI around 0.58, sandstone- and clay-dominated strata, proximity within 10 km of faults, 60-minute rainfall of about 40 mm, and seismic acceleration near 0.05 g. The susceptibility maps generated by the four models exhibit strong spatial consistency and highlight elevated landslide risk along pipelines in eastern Sichuan. It is concluded that the proposed hybrid transfer learning framework effectively addresses data limitations and negative sample uncertainty, providing accurate and robust susceptibility predictions and offering practical decision support for targeted inspection and hazard mitigation of pipeline infrastructure.