Abstract: With the construction of large-scale radio telescopes, detecting pulsars from large-scale pulse signals has become an important task of space exploration. Machine learning algorithms are favored in single-pulse data analysis due to their data-driven advantages. However, algorithms used in pulsar searching cannot guarantee that their results are global optimal solutions. In this paper, eXtreme Gradient Boosting (XGBoost) method is studied in single pulse classification with the data from the LOFAR Tied-Array All-Sky Survey (LOTAAS). The LOTAAS is an ongoing survey of the Northern sky for pulsars and transients with LOFAR using a digital aperture array. In January 2019, the LOTAAS survey has discovered and confirmed 73 radio pulsars, which demonstrates its ability to find new pulsars. A fully labeled data set used for training and validation of the machine model is necessary. However, faced with massive amounts of astronomical observation data, it's time-consuming and laborious work to labeling data with manual inspection. In this study, we directly use the well-prepared data in the work of Michilli et al. (2018) for saving the labor of repetitive processing of data. In order to verify the performance of XGBoost method, this paper compares the algorithm with other four machine learning models. The results show that XGBoost combined with wrapped feature selection method has more advantages in single pulse recognition, with the lowest misclassification rate and the highest accuracy, and F₁ score. This study has important implications for pulsar monitoring and can provide a reference for the research of single pulse search based on Five-hundred-meter Aperture Spherical radio Telescope (FAST) signals in China.