normalize each dimension to be zero mean and unit variance.. That is, normalizing the features changes the prediction made by the kNN model. Is kNN still sensitive to data normalization if cosine similarity is used as the distance metric? If your answer is "yes", show that this is the case using a simple example (i.e., create a data set and demonstrate the difference). If your answer is "no", prove it. At some point, it might be come helpful to think geometrically rather than algebraically. Cosine similarity is a measure of angle distance. How do angles behave when features are scaled? You might come up with a rough sketch in your mind before trying to plug in some numbers that work