Interactive explainer
k-means clustering – interactive calculator
Which customers are similar? The answer also depends on how similarity is measured. Group the same synthetic customers by order count and monthly revenue, and explore the effect of scaling.
Group 90 synthetic customers using monthly order count and revenue. Choose two to five groups, toggle standardisation and change the internal revenue unit. The original observations and starting points stay fixed for a given group count. All calculations run locally; no real customer data is required.
k-means clustering – interactive calculator
k-means assigns each point to its nearest centre and recalculates group means. The chart always shows original order counts and euros, regardless of the internal scaling choice.
Three experiments
- Switch units without standardisation. Turn standardisation off and compare euros with thousands of euros. Check group sizes and centres in the table. Assignments can differ even though exactly the same customers are being analysed.
- Remove the unit effect. Enable standardisation and repeat the switch. Each feature’s mean and standard deviation are taken into account. Standardised distances and assignments therefore stay the same apart from numerical rounding.
- Allow more groups. Compare two, three and five groups. For each partition, consider what specific action it would support. More detailed groups are not automatically more useful. Use Reset to return to the starting configuration.
What you can conclude
Groups express proximity in the chosen feature space, not established customer types. Fixed starting points make comparisons reproducible but do not guarantee a global optimum. Outliers, unequal spreads and non-compact structures can change the result. Centres are always reported in the original units. The clustering article explains the distance calculation and why standardisation remains a substantive modelling choice.