I have categorical data. For some variables there are two levels/categories. For other number of levels or categories are more than 2. I want to apply K-modes clustering using R. Will someone please guide how I can choose optimum number of clusters? I came across similar posts/questions but not clear .
Regards
Hi Tariq
The different numbers of levels is not a problem as they are not too large (e.g. more than 40 or 50 levels). The clustering is an empiric process and usually we process one k-clustering then observe the results. You may process others clustering with a different k if the results have no meaning for you but also to compare the different results and see if one constant structure appears.
R also provide useful packages : clusterR with the Optimal_Clusters_KMeans function and NbClust for determining the best number of clusters.
Good luck :)
Claire Tissot
I suggest you follow https://www.datanovia.com/en/lessons/determining-the-optimal-number-of-clusters-3-must-know-methods/
https://towardsdatascience.com/10-tips-for-choosing-the-optimal-number-of-clusters-277e93d72d92
Also https://medium.com/analytics-vidhya/how-to-determine-the-optimal-k-for-k-means-708505d204eb
There is something called the Calinski-Harabasz Criterion which is used to determine the optimum number of clusters.
https://in.mathworks.com/help/stats/clustering.evaluation.calinskiharabaszevaluation-class.html
You may use a combination of FAMD and HCPC (consolidated) functions of R package FactoMineR to get a recommendation.
N.
- Badges
- Science topic
- Similar topics
- Psychology