In order to solve the problem of complex and low accuracy of motor drive fault diagnosis in automatic packaging production line, and to improve the stability of motor operation and personnel safety in complex production environment, a precise prediction method of motor drive fault based on XGBoost feature reconstruction and neural network prediction is proposed. The method first uses a part of the training data to construct a feature tree through the XGBoost algorithm, and then inputs the remaining training data into the XGBoost algorithm to obtain the reconstructed features. Consequently, using the One-hot encoding to map the reconstructed features to the Euclidean space to further amplify the difference in features. Finally, the obtained features are input into the neural network model with parameter adjustment to complete the fault prediction. Compared with the neural network model constructed without XGBoost features, the structure proposed in this paper achieves nearly 100% classification accuracy on the verification set and the test set of the data test set randomly divided, which verifies the effectiveness and stability of the model. The sensorless high-precision diagnosis of the motor drive fault in automatic packaging production line is realized, which is beneficial to improve motor stability and personnel safety in complex production environment.