Feature Importance is negative, while accuracy of the classification model is above 90%

Question

I have built a classification model in predicting the churn of the user. The feature importance is coming negative for almost all the variables but the accuracy of the model