Benchmark : XgBoost performance on fashion-mnist 89.8% and on MNIST 96.8%

[anktplwl91]

Hello, I tried XgBoost on both Fashion-MNIST and MNIST dataset, with the only pre-processing as scaling the pixel values to mean=0.0 and var=1.0.

Fashion-MNIST
Train accuracy 99.5%
Validation accuracy 90.7%
Test accuracy 89.8%

MNIST
Train accuracy 99.7%
Validation accuracy 97.4%
Test accuracy 96.8%

Notebook link
https://github.com/anktplwl91/fashion_mnist.git

[SynthaxWarrior]

Isn't it wrong to do fit_transform() instead of transform() on test set while scaling?