sandstone and mudstone identification tasks are usually based on logging curves and rely on traditional methods such as empirical formulas，field core sampling，cross plots，and cluster analysis，but these methods fail to make full use of the sandstone and mudstone features contained in the logging curves. At the same time，these traditional methods have low accuracy and slow efficiency and are greatly affected by human factors. To address the above problems，this paper uses logging data as the basis，combines the key technical difficulties of sandstone and mudstone identification, and conducts sensitivity analysis on logging parameters，so as to select appropriate influencing factors and construct a complete training data set through several pre-processing operations. In addition，the paper introduces the Focal Loss function and proposes the FL-XGBoost model according to the sparsity of logging labels and carries out sandstone and mudstone identification in Niuzhuang area of Shengli Oilfield. The experimental results show that the sandstone and mudstone identification model using the FL-XGBoost algorithm achieves an accuracy of 0.827 in identifying the sandstone and mudstone in the study area. Finally，the strong generalization ability of the FL-XGBoost algorithm in the identification classification field is verified through five publicly classified dataset design comparison experiments.