This study aims to solve the problems of insufficient accuracy and low efficiency of the existing methods in sprint pattern recognition to optimize the training and competition strategies of athletes. Firstly, the data collected in this study come from high-precision sensors and computer simulation, involving key biomechanical parameters in sprint, such as step frequency, stride length and acceleration. The dataset covers multiple tests of multiple athletes, ensuring the diversity of samples. Secondly, an optimized machine learning algorithm based on decision tree is adopted. It combines the advantages of Random Forest (RF) and Gradient Boosting Tree (GBT), and improves the accuracy and efficiency of the model in sprint pattern recognition by adaptively adjusting the hyperparameter and tree structure. Specifically, by introducing adaptive feature selection and ensemble learning methods, the decision tree algorithm effectively improves the recognition ability of the model for different athletes and sports states, thus reducing the over-fitting phenomenon and improving the generalization ability. In the process of model training, cross-validation and grid search optimization methods are adopted to ensure the reasonable selection of super parameters. Moreover, the superiority of the model is verified by comparing with the commonly used algorithms such as Support Vector Machine (SVM) and Convolutional Neural Network (CNN). The accuracy rate on the test set is 94.9%, which is higher than that of SVM (87.0%) and CNN (92.0%). In addition, the optimized decision tree algorithm performs well in computational efficiency. However, the training data of this model comes from the simulation environment, which may deviate from the real game data. Future research can verify the generalization ability of the model through more actual data.