Overfitting issues.

[Luab]

trafficstars

Good day. Thanks for sharing your code. I have noticed an error in your train/test splitting. First, you create overlapping windows and then divide data into train/test parts. This means that same data points go to both train and test distributions, resulting in significant overfit of your models.

[tongdaxu]

Crap, you are correct about the overlapping windows. Random Forest does not over-fit but the accuracy lows down to 72%.