Binary Cross Entropy Loss

After preprocessing my data. I have applied XG gradient boosting to fit my model after two or three epochs my cost is becoming constant and not reducing to a significant number. Why this is happening?