Davies-Bouldin Index
Davies-Bouldin Index for Cluster Separation · Also known as: DBI, Davies Bouldin index
The Davies-Bouldin Index, introduced by Davies and Bouldin in 1979, is a metric for evaluating clustering quality based on the average similarity between each cluster and its most similar neighboring cluster. Lower values indicate better clustering, with a minimum of 0 representing perfectly separated, non-overlapping clusters.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use the Davies-Bouldin Index when you need an easily interpretable cluster quality metric that emphasizes both compactness and separation. It requires no external labels and works with any distance metric and clustering algorithm. However, it assumes convex clusters and is sensitive to the number of clusters; it tends to favor solutions with more clusters.
Strengths & limitations
- Simple to understand: lower values always indicate better clustering
- Works with any distance metric and clustering algorithm
- No external labels required; purely unsupervised evaluation
- Computationally efficient compared to some other metrics
- Provides both global and cluster-level insights
- Biased toward convex, spherical clusters
- Tends to increase with the number of clusters, making it unsuitable for selecting k without modification
- Assumes similar cluster sizes; fails on clusters with very different densities or sizes
- Less interpretable for high-dimensional data
Frequently asked
What is a good Davies-Bouldin Index value?
Lower is better; an index value less than 1.0 generally indicates well-separated clusters, values between 1.0 and 1.5 suggest reasonable clustering, and values above 1.5 indicate poor cluster separation. However, interpretation depends on your data and domain.
Why does the Davies-Bouldin Index increase with more clusters?
The index measures the average maximum similarity ratio for each cluster. As you add clusters, points spread out more and clusters shrink, leading to lower within-cluster distances and lower index values by definition. This bias should be corrected for when using DBI for cluster number selection.
How sensitive is the Davies-Bouldin Index to outliers?
The index can be sensitive to outliers since it uses centroid-based distance calculations. Outliers shift centroids and inflate within-cluster averages, potentially distorting the metric. Preprocess data to handle outliers before applying the index.
Can I use the Davies-Bouldin Index for comparing different clustering algorithms?
Yes; a lower DBI indicates better performance. However, ensure you use the same distance metric and data preprocessing across algorithms for fair comparison.
Sources
- Davies, D. L., & Bouldin, D. W. (1979). A cluster separation measure. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1(2), 224-227. DOI: 10.1109/TPAMI.1979.4766909 ↗
How to cite this page
ScholarGate. (2026, June 3). Davies-Bouldin Index for Cluster Separation. ScholarGate. https://scholargate.app/en/model-evaluation/davies-bouldin-index
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Adjusted Rand IndexModel Evaluation↔ compare
- Calinski-Harabasz IndexModel Evaluation↔ compare
- Dunn IndexModel Evaluation↔ compare
- Gap StatisticModel Evaluation↔ compare
- Silhouette ScoreModel Evaluation↔ compare