for shure machine learning approach have limitations and problems when dealing with financial data... but in respect of the construction of my data set to train the algorithm I dont have any data leakage or future being used on the predictions... if you could elaborate more about the problems with this approach, I would love to read about.