Orange: Cluster Inspection
https://orange.biolab.si/workflows/
Kita menggunakan zoo data set dalam kombinasi dengan widget Hierarchical Clustering untuk menemukan kelompok hewan. Sekarang kita memiliki cluster yang ingin kita temukan dan apa yang signifikan untuk setiap cluster! Berikan cluster ke widget Box Plot dan gunakan ‘Order by relevance’ untuk menemukan apa yang mendefinisikan cluster. Tampaknya mereka dipisahkan oleh jenisnya, meskipun pengelompokannya dilakukan tanpa ada class label! Ini adalah contoh unsupervised learning.
The workflow depicted in the Orange Data Mining tool follows a Hierarchical Clustering approach using a zoo dataset of 101 animals described by 16 attributes. Here’s the step-by-step breakdown:
1. File (Data Loading)
- The process starts with loading the zoo dataset (101 animals, 16 attributes) using the File widget.
2. Distances (Pairwise Similarity Computation)
- Next, the Distances widget computes pairwise similarities (distances) between animals.
- It uses the Euclidean distance to measure how similar or different the animals are based on their attributes.
3. Hierarchical Clustering (Grouping the Animals into Clusters)
- The computed distances are passed to the Hierarchical Clustering widget.
- The desired number of clusters is set (recommended between 4 to 6 clusters).
- Ward linkage is used as the clustering method because it works best for this type of data.
4. Box Plot (Cluster Inspection & Analysis)
- The selected clusters are inspected using the Box Plot widget.
- Users can set the variable to Type and Subgroups to Cluster to analyze how the clustering aligns with different animal types.
- This allows for visualizing differences between clusters based on attributes.
Summary
This workflow in Orange Data Mining starts with loading the dataset, computes distances between data points, applies hierarchical clustering to group them, and finally analyzes the resulting clusters through box plots. This approach is useful for understanding similarities among animals based on given attributes.
Source