Maybe feature engineering. ( I only have an initial background on machine learning stuff. But I know that if I want to calculate the relation between a possible intervention and a trouble, a good way is to use statists and knowledge related to machine learning )