In this article, we delve into the comparative analysis of K-Means and Hierarchical Clustering techniques. As part of Task 19 of the Bytewise Fellowship, this project examines how these clustering methods perform across different datasets, including Iris and Mall Customers. We explore how the characteristics of each dataset influence clustering outcomes and provide a detailed comparison of these popular algorithms.
Dataset Selection
- Iris Dataset: Contains measurements of iris flowers.
- Mall Customers Dataset: Includes customer data like annual income and spending score.
Initial EDA
- Data Distribution: Analyzed feature distribution and outliers.
- Feature Correlations: Examined feature relationships.
K-Means Clustering
- Application: Applied K-Means and determined optimal clusters using the Elbow Method and Silhouette Score.
Hierarchical Clustering
- Application: Implemented with various linkage criteria and visualized using dendrograms.
Visualization
- Dimensionality Reduction: Used PCA and t-SNE to visualize clusters.
- Comparison: Evaluated cluster compactness and separation.
Cluster Analysis
- Iris Dataset: Identified species groups.
- Mall Customers Dataset: Discovered customer segments.
Impact of Parameters
- K-Means: Examined effects of different cluster numbers.
- Hierarchical: Analyzed effects of linkage criteria.
Effectiveness
- Comparative Analysis: Compared K-Means and Hierarchical Clustering for each dataset, discussing performance and suitability.
This analysis reveals how different clustering algorithms can be more or less effective depending on dataset characteristics. K-Means excels with clear, spherical clusters, while Hierarchical Clustering provides flexibility for hierarchical data. The study emphasizes the importance of choosing the appropriate algorithm based on data distribution and characteristics.