It's hard to imagine if once can ever avoid assumption violations. If we are using KMeans with Euclidean distance, it is highly likely to have correlated features. Yet, we still use Euclidean distance because we don't want to create new features that are not interpretable. So there is always that tradeoff between interpretability and accuracy that you would have to consider.
How do we use other clustering methods as KNN primarily uses Euclidean distances ?
It's hard to imagine if once can ever avoid assumption violations. If we are using KMeans with Euclidean distance, it is highly likely to have correlated features. Yet, we still use Euclidean distance because we don't want to create new features that are not interpretable. So there is always that tradeoff between interpretability and accuracy that you would have to consider.
FYI, I really look forward to receiving your daily updates! They are filled with usable nuggets of info. THANK YOU!
Thanks for appreciating, Teddy. Happy to know you are loving these daily insights :)
Great read