I would argue that it would often be easier to implement a ML solution on regular data than try and clean the data and then use linear regression