This document details an assignment where three different clustering techniques were applied to a dataset. The performance of these clustering algorithms was evaluated across various preprocessing methods and different numbers of clusters.
The three clustering methods employed are:
- KMeans Clustering
- Hierarchical Clustering
- MeanShift Clustering
The dataset utilized is the Anuran Calls (MFCCs) dataset sourced from the UCI dataset library. This dataset comprises acoustic features extracted from syllables of frog calls, including labels for family, genus, and species.
- Number of Instances: 7195
- Number of Attributes: 25 + 1 (ID)
Originally created for tasks related to anuran species recognition through their calls, this dataset is multilabel and consists of three label columns. It comprises segments of 60 audio records from 4 different families, 8 genera, and 10 species. Each audio record corresponds to an individual frog specimen, with an additional column for record ID.
Three metrics were employed to assess the performance of clustering algorithms:
-
Silhouette Score: Evaluates the clustering algorithm's performance based on the compactness of individual clusters (intra-cluster distance) and separation among clusters (inter-cluster distance). The score is calculated per sample and then averaged across all samples.
-
Calinski-Harabasz Score: Defined as the ratio of the sum of between-cluster dispersion to within-cluster dispersion.
-
Davies-Bouldin Index: A validation metric measuring the average similarity between each cluster and the cluster most similar to it. It assesses clusters based on the ratio between inter-cluster and intra-cluster distances.
Results for each clustering method are displayed below: