@Venkatesh's answer is not right. Logistic Regression uses gradient-descent based optimization, which will be sensitive to the scale of the features: if a feature lives on a much larger scale, its corresponding element in the gradient will be much larger, and the optimization will favor going "downhill" in the direction of this feature. This can be mitigated by having appropriately scaled learning rates for each feature (scaled by the std of the feature, specifically), but I don't think that most off-the shelf logistic regression APIs do that.

The part about tree-based algorithm is right though.