⚠️ Data Leakage: Must not use test data when fitting MinMaxScaler()

[shure-dev]

Probably, I found a serious error.

If I'm correct, we cannot use any information from test data when preprocessing data.

However, your code applied fit_transform() to train and test data.

This means train data can contain information from test data and effects accuracy.

Please correct me if my idea is wrong, thank you.

[shure-dev]

This answer seems working well for this issue.

https://stackoverflow.com/questions/70923839/sklearn-preprocessing-with-a-rolling-window

[shure-dev]

Probably, also we have to care about stationarity, when we treat time series data