Cluster evaluation metrics in ML Python - Model Pipeline Trace

Model Pipeline - Cluster evaluation metrics

This pipeline shows how clustering groups data points and how we measure the quality of these groups using cluster evaluation metrics.

Data Flow - 5 Stages

1Data in

500 rows x 4 columns→Raw dataset with features→500 rows x 4 columns

[5.1, 3.5, 1.4, 0.2]

↓

2Preprocessing

500 rows x 4 columns→Scaling features to range 0-1→500 rows x 4 columns

[0.22, 0.58, 0.15, 0.10]

↓

3Feature Engineering

500 rows x 4 columns→No new features added, data ready for clustering→500 rows x 4 columns

[0.22, 0.58, 0.15, 0.10]

↓

4Model Trains

500 rows x 4 columns→K-Means clustering with k=3→500 rows x 1 cluster label

[0, 2, 1, 0, 2]

↓

5Metrics Compute

500 rows x 4 columns and 500 rows x 1 cluster label→Calculate Silhouette Score, Davies-Bouldin Index, and Calinski-Harabasz Index→3 scalar metric values

Silhouette Score=0.55, Davies-Bouldin=0.75, Calinski-Harabasz=320.4

Training Trace - Epoch by Epoch

N/A

Epoch	Loss ↓	Accuracy ↑	Observation
1	N/A	N/A	Clustering is unsupervised; no loss or accuracy. Initial cluster centers chosen.

Prediction Trace - 4 Layers

Layer 1: Input sample

Layer 2: Distance calculation

Layer 3: Assign cluster label

Layer 4: Silhouette score calculation

Model Quiz - 3 Questions

What does a higher Silhouette Score indicate about clusters?

A Clusters are well separated and dense

B Clusters overlap a lot

C Clusters have many outliers

D Clusters have fewer samples

Key Insight

Cluster evaluation metrics help us understand how good our groups are without needing labels. Silhouette Score, Davies-Bouldin, and Calinski-Harabasz each give a different view of cluster quality.