Feature scaling pros and cons

I’ve decided to scale 2 features cause they have values in range 50+
It turns out that model accuracy before scaling is higher (0.8) than after scaling (0.75).

Does scaling always improve model accuracy and this is my problem or we should avoid scaling when it performs bad results?

Before scaling plot

After scaling plot

The main benefit of feature scaling and normalization is that it makes gradient descent work better - you can get a better solution faster because you can use a higher learning rate and fewer iterations.