E.g. I have a given interest of a balance between quality, cost and time. I have in the data set 6 features: 3 features for cost, 2 for time and only 1 for quality.
I would like to:
- quality features * 2
- cost features *2/3
- times *1
When should I do that: before the transformation/scaling? after the transformation but before scaling or even after trasnfromation/scaling? and why then?
-Data set is very small:
Dear Isabelle Hacker
Before clustering, it is customary practice to normalize all variables. It makes no difference whether you use full linkage vs. any other linkage, or hierarchical clustering vs. another method (e.g., k-means).
Clustering models are distance-based algorithms that employ a distance metric to detect similarities between data and construct clusters. As a result, characteristics with large ranges will have a greater effect on clustering. As a result, prior to developing a clustering model, standardization is essential.
The clustering properties, like those employed in the k-NN approach, must be quantified in equivalent units. Normalization or standardization are not required.
Kind Regards
Qamar Ul Islam
Thanks, that refers to a general clustering but what about "When do I need to weight my features before e.g. a hierarchical clustering or kmeans?"
- Badges
- Science topic