Data leakage when setting class_weight to tackle imbalanced time series data?

Question

I\'m using a random forest classifier to predict whether a stock\'s return for the next period is greater than a certain threshold (say -2%), so negative is 0 and positive i