[python] When performing k-means clustering, normalize and standardize the data.

Overview

While working on a machine learning problem, I tried k-means clustering of the data. However, it seemed that it depended too much on the value of one column. To prevent this from happening, the size of the data must be the same in advance.

What is k-means clustering

k-means clustering is an algorithm that divides data into multiple clusters. Create a number of cluster centroids specified in advance and classify each data into the nearest cluster.

Why normalization is necessary

The distance between data points in the coordinate space is used in cluster classification in k-means. Therefore, if the absolute value of only one piece of data is large, the distance between points will be large even with the same variance, which will affect the calculation of cluster classification.

Even with the same 10% difference, the distance between points is significantly different.

The data I was actually using was such that columns A, C, and D ranged from 0 to 2, and column B ranged from 1000 to 3000. When I forgot to standardize and tried clustering, the clustering seemed to be affected only by column B. I'm glad I noticed something was wrong this time.

Colclusion

Be sure to normalize and standardize before performing kmeans clustering. And be sure to check the result of clustering.