There are many factors that affect the faults sealing capacity, but the traditional lateral sealing model only considers the clay content, which leads to the great limitations of the established evaluation model. The objective of this study is to comprehensively examine the influence of potentially contributing factors on the fault sealing capacity. The Shuangtaizi structure is selected as the research focus in this study, where a 3D geological model is established to conduct detailed reservoir characterization for calculating trap elements and hydrocarbon column height. Subsequently, a dataset on the sealing attributes of fault-bounded traps is constructed after optimizing the dataset. By utilizing both machine learning evaluation models and traditional methods, the evaluation of the lateral sealing capacity of Shuangtaizi structural faults is performed. The influencing factors of fault rock sealing are clearly delineated as ten key parameters: effective normal stress, clay content, fault strike, fault throw, dip angle, transverse gradient, dip slip gradient, longitudinal gradient, longitudinal strain and shear strain in descending order of importance. Among these factors analyzed for fault sealing considerations, the most significant is the effective normal stress followed by clay content. By establishing a novel machine learning-based lateral sealing evaluation model and comparing it with the traditional SGR-AFPD fault lateral sealing evaluation model, it could be observed that the machine learning model exhibits a smaller error range and higher confidence in its evaluation results compared to the traditional model. Notably, the conventional model fails to consider the crucial influence of effective normal stress on sealing performance, which accounts for its inadequate accuracy. The model incorporates a comprehensive assessment of various factors that influence the lateral sealing capability of faults. In order to obtain objective and more realistic evaluation results, it is imperative to establish a multidimensional evaluation model in future studies.