Hi,

The error for each value of k is not necessarily always going to strictly decrease, it depends on the structure of your data and the locations of the initial centroids chosen in the algorithm. You can see in the chart that the error also goes up a bit from 9 to 10 clusters, from 15 to 16, etc., not just from 7 to 8. When I look at a chart like this, I am more focused on the larger trends, like the move from 2 to 3 clusters.