Metric for hyperparamter search in Unsupervise Learning
Metrics
Clustering is an unsupervised machine learning technique used to group similar data points together. Here are 3 common evaluation metrics for clustering algorithms:
-
Silhouette Score: This metric calculates the similarity of each data point to its own cluster compared to other clusters. The score ranges from -1 to 1, with higher scores indicating better cluster quality.
-
Davies-Bouldin Index: This metric measures the average similarity between each cluster and its most similar cluster. The goal is to minimize this metric, as it indicates well-separated and distinct clusters.
-
Within Cluster Sum of Squares (WCSS): This metric measures the sum of squared distances between each data point and its centroid within a cluster. The goal is to minimize this metric, as it indicates that the data points are tightly clustered around their centroids.
Comparison
Davies-Bouldin Index (DBI) and Silhouette Score measure the quality of clusters in terms of their separation and compactness, while Within Cluster Sum of Squares (WCSS) measures the compactness of each cluster.
WCSS measures the sum of squared distances of each data point to its centroid within the cluster. It is used to determine the optimal number of clusters for a given dataset. The goal is to minimize the WCSS, which means the data points in each cluster are closer to their centroid. However, WCSS does not measure the separation between the clusters.
DBI measures the average similarity between each cluster and its most similar cluster, normalized by the average distance between the clusters and their centroids. It is used to evaluate the separation and compactness of the clusters. The goal is to minimize the DBI, which means the clusters are well-separated and compact.
Silhouette Score measures the similarity of each data point to its own cluster compared to other clusters. The score ranges from -1 to 1, with higher scores indicating better cluster quality. Silhouette Score takes into account both the cohesion within each cluster and the separation between clusters. A high Silhouette Score indicates that the clusters are well-separated and dense, while a low Silhouette Score indicates that the clusters are overlapping or poorly separated.
Conclusion
In general, DBI and Silhouette Score are more suitable for evaluating the quality of the clustering results in terms of their separation and compactness, while WCSS is more suitable for determining the optimal number of clusters. However, it is recommended to use all three metrics in combination to get a comprehensive understanding of the clustering algorithm’s performance. A good clustering algorithm should produce clusters with low WCSS, low DBI, and high Silhouette Score.