There are many kinds of swing actions in table tennis,and accurately identifying these actions is of great significance to swing analysis.For this end, this paper designs a swing action recognition scheme based on the MPU9250 sensor.Firstly, the raw data collected by the sensor is processed using the sliding window and filtering.After that, twentytwo statistical features, such as mean, variance, period and so on, are extracted by performing timedomain analysis on the preprocessed data.Secondly, an embedded feature selection strategy is exploited for feature filtering, and subsequently generates ten distinguished features.Then support vector machine, random forest, decision tree and an ensemble learning based on above three algorithms are used for classifier training.Lastly, the four classifiers are fully tested for the recognition performance of four motion modes such as forehand attack, backhand push, forehand rub and backhand rub.The results showed that the ensemble learning performed best with an average recognition accuracy of 94.25%.