Evaluating Cluster Analysis in Python: Using DBSCAN and Validity Indices

Welcome to our Cluster Performance Unveiled course lesson! Here, we leverage Silhouette Scores, the Davies-Bouldin Index, and Cross-Tabulation Analysis to assess DBSCAN, a top-performing clustering algorithm with a focus on density. Exciting, right?

DBSCAN has advantages when the number of clusters is undetermined and density plays a key role in the formation of clusters. Using Python’s sklearn library, executing the DBSCAN algorithm is simple.

from sklearn.cluster import DBSCAN
import numpy as np

# Our dataset
X = np.array([[1, 2], [2, 2], [2, 3], [8, 7], [8, 8], [25, 80]])

# Setup the DBSCAN model with eps and min_samples parameters
dbscan = DBSCAN(eps=3, min_samples=2)

# Fit the model to our dataset
dbscan.fit(X)

We implement DBSCAN with eps and min_samples parameters, which denote the maximum distance between neighbor points and the sample count for a point to be a core point, respectively. After fitting our algorithm, we need a quantitative assessment of how well the clustering performed. The Silhouette Score works as a solid indicator of cluster quality, capturing the mean intra-cluster distance (a) and the mean nearest-cluster distance (b) for each sample. It then subtracts the mean distance within the cluster (a) from the mean distance to the nearest cluster (b) and calculates their ratio. It's closer to 1 when the clusters are dense and well-separated.

from sklearn.metrics import silhouette_score

labels = dbscan.labels_  # Get the cluster labels from our DBSCAN model

# Compute and print Silhouette Score
score = silhouette_score(X, labels)

print(f"Silhouette Score: {score}")

Remarkably, a high score signifies that data points form well-defined clusters.

The Davies-Bouldin Index plays a crucial role in evaluating the quality of clustering models. It computes the average measure of similarity between each cluster and its most similar cluster, with lower values suggesting better partitioning. It's calculated as the ratio of within-cluster distances to between-cluster distances.

from sklearn.metrics import davies_bouldin_score

# Compute and print Davies-Bouldin Index
db = davies_bouldin_score(X, labels)

print(f"Davies-Bouldin Index: {db}")

A lower Davies-Bouldin Index is desirable, as it hints at better cluster separation. In addition, we can further evaluate our model by performing a Cross-Tabulation Analysis.

import pandas as pd

# Assuming `true_labels` has the true labels for our data points
cross_tab = pd.crosstab(labels, true_labels)

print(cross_tab)

Cross-Tabulation Analysis generates a matrix, providing a comparison of the model's performance against the actual labels.