A comprehensive comparison of K-Means and Mini-Batch K-Means clustering algorithms on air quality data, featuring automatic parameter tuning and detailed performance analysis.
This project performs a head-to-head comparison between standard K-Means and Mini-Batch K-Means clustering algorithms. It automatically evaluates cluster numbers from k=2 to k=10, using both the Elbow Method and Silhouette Score to determine optimal clustering configurations.
Features analyzed:
- PM10 (pm_sepuluh)
- PM2.5 (pm_duakomalima)
- Sulfur Dioxide (sulfur_dioksida)
- Carbon Monoxide (karbon_monoksida)
- Ozone (ozon)
- Nitrogen Dioxide (nitrogen_dioksida)
- Automated parameter tuning (k=2 to k=10)
- Dual algorithm comparison (K-Means vs Mini-Batch K-Means)
- Multiple evaluation metrics (WCSS, Silhouette Score, Execution Time)
- PCA visualization for cluster interpretation
- Smart sampling for large datasets (>10k rows)
- Comprehensive reporting (CSV, Excel, PNG outputs)
Python 3.7+pip install pandas numpy matplotlib seaborn scikit-learn openpyxlOr install from requirements.txt:
pip install -r requirements.txt- Run the analysis:
python main.py- Check the
output/folder for results
The Elbow Method helps identify the optimal number of clusters by plotting WCSS (Within-Cluster Sum of Squares) against different k values. The "elbow point" indicates where adding more clusters provides diminishing returns.
Silhouette Score measures how well each data point fits within its assigned cluster. Higher scores indicate better-defined clusters. This comparison shows the performance difference between K-Means and Mini-Batch K-Means across different k values.
PCA (Principal Component Analysis) reduces the 6-dimensional feature space to 2D for visualization. This plot shows how K-Means groups the air quality data into distinct clusters.
Similar to the K-Means visualization, but using the Mini-Batch algorithm which is optimized for larger datasets through batch processing.
After running the analysis, the following files are generated in the output/ folder:
| File | Description |
|---|---|
comparison_auto_elbow_method.png |
Elbow method comparison chart |
comparison_auto_silhouette_score.png |
Silhouette score comparison chart |
comparison_auto_pca_kmeans_k*.png |
K-Means PCA visualization |
comparison_auto_pca_minibatch_k*.png |
Mini-Batch K-Means PCA visualization |
hasil_tabel_perbandingan_tuning.csv |
Complete performance metrics table |
comparison_auto_profil_klaster.xlsx |
Cluster profiles with centroids |
comparison_auto_data_labeled.csv |
Original data with cluster labels |
The hasil_tabel_perbandingan_tuning.csv contains:
- WCSS values for both algorithms
- Silhouette scores for both algorithms
- Execution time comparisons
- Optimal k recommendations
The comparison_auto_profil_klaster.xlsx contains:
- Sheet 1: K-Means cluster centroids and member counts
- Sheet 2: Mini-Batch K-Means cluster centroids and member counts
- Pros: More accurate cluster assignments
- Cons: Slower on large datasets
- Best for: Smaller datasets (<10k rows) where accuracy is priority
- Pros: Much faster on large datasets
- Cons: Slightly less accurate than standard K-Means
- Best for: Large datasets (>10k rows) where speed matters
- Lower is better
- Look for the "elbow" where the rate of decrease slows
- Typical range: Varies by dataset scale
- Range: -1 to 1
- >0.5: Good clustering
- 0.25-0.5: Reasonable clustering
- <0.25: Poor clustering
- Standard K-Means is typically 2-5x slower than Mini-Batch
- Gap widens with larger datasets
- Scikit-learn for clustering implementations
- Matplotlib & Seaborn for visualizations
- Pandas for data manipulation