I'm surprised no one else has commented that a few of the conceptual comments in this article are a bit odd or just wrong. > The final accuracy is 90% because 1 of the 10 observations is on the incorrect side of the decision boundary. Who is using K-means for classification? If you have labels, then a supervised algorithm seems like a more appropriate choice. > K-means clustering is a recursive algorithm It is? > If…
> K-means clustering is a recursive algorithm My bad. It's iterative. I'll fix that. Thanks.
> If we know that the distributions are Gaussian, which is very frequently the case in machine learning Gaussian distributions are very frequent and important in machine learning because of the Central Limit Theorem but, beyond that, you are correct. While many natural phenomena are approximately normal, the reason for the Gaussian's frequent use is often mathematical mathematical convenience. I'll correct my post.
> we can employ a more powerful algorithm: Expectation Maximization (EM) Excellent point. I will fix that, too. "While k-means is simple, it does not take advantage of our knowledge of the Gaussian nature of the data. If we know that the distributions are at least approximately Gaussian, which is frequently the case, we can employ a more powerful application of the Expectation Maximization (EM) framework (k-means is a specific implementation of centroid-based clustering that uses an iterative approach similar to EM with 'hard' clustering) that takes advantage of this." Thank you for pointing out all of this!