Samenvatting Data Analytics & AI (pt1) | Clustering & Dimensionality Reduction | KU Leuven | 2025/26

Data & AI pt. 1 (methodes)


CLUSTERING

Hard clustering (elk datapunt behoort maar tot exact 1 cluster)

1. Hierarchical clustering  dendrogram
 Agglomerative (bottom-up):
o Starts with each single datapoint as 1 cluster
o Starts combining them until you have 1 big cluster with ALL
datapoints
 Divisive (top-down):
o Starts with 1 big cluster
o Divides them in smaller clusters until every datapoint is 1
single cluster

1.1. Single linkage: adds 2 clusters/datapoints together when the
minimal distance between 2 points (1 from each) is the smallest
- Clusters A, B, C: kleinste afstand is tussen A & B  die
samenvoegen
- Ketens van clusters
1.2. Complete linkage: adds 2 clusters/datapoints together when
the maximal distance between 2 points (1 from each) is the
smallest
- Clusters A, B, C: kleinste maximale afstand is tussen A & C  die
samenvoegen
- Compacte clusters
1.3. Average linkage: adds 2 clusters/datapoints together when the
average distance between 2 points (1 from each) is the smallest
- Clusters A, B, C: kleinste average afstand is tussen B & C  die
samenvoegen
1.4. Ward method: 2 clusters are added together when the new
(bigger) cluster leads to the smallest increase in within-cluster sum
of squares
- WCSS = hoe ver punten uit elkaar liggen binnen dezelfde cluster
o Kleiner = dichter bij elkaar, dichter bij centroid (minder
variatie in de punten binnen die cluster)
- Ronde clusters
 Dia 7




1

, 2. Partition clustering
2.1. K-means clustering: euclidean
- For numerical
- Algoritme (dia 13)
- Standard method
2.2. K-means ++: euclidean
- For numerical
- Algoritme (dia 15)
2.3. K-mode: hamming distance
- For categorical data
- Algoritme (dia 18)
2.4. K-prototype: Euclidean & hamming
- For numerical & categorical
- Algoritme (dia 21)

3. HDBSCAN (density based) clustering
- Algoritme (dia 34)



Soft clustering (datapunt kan tot meerdere clusters behoren, met een
bepaalde kans)

1. Fuzzy c-means (FCM)
- Algoritme (dia 59)

2. Gaussian mixture model (GMM)
 Gaussian & independence assumptions!!! (dia 74)
- Algoritme (dia 75)

VERGELIJKING VAN DE 2 = DIA 85




2