/TimeSeries-Clustering

Primary LanguageJupyter Notebook

TimeSeries-Clustering

This project demonstrates various clustering methods and visualizations for time series data. It includes utility functions for running clustering algorithms, plotting results, and performing silhouette analysis.

Team Members

  • Abdesselam BENAMEUR
  • Hakim IGUENI
  • Salma TALANTIKITE

Contents

  • utils.py: Contains utility functions for clustering and visualization.
  • ProjetUL.ipynb: Jupyter notebook demonstrating the use of the utility functions on a dataset.

Setup

Requirements

Make sure you have the following Python packages installed:

  • numpy
  • pandas
  • matplotlib
  • seaborn
  • scipy
  • scikit-learn
  • tslearn
  • kmedoids

You can install these packages using pip:

pip install numpy pandas matplotlib seaborn scipy scikit-learn tslearn kmedoids

Data

The data used in this project is loaded from the following URLs:

Usage

Utility Functions

The utils.py file contains the following utility functions:

  • run_clustering_method(clustering_method, X, metric='precomputed', verbose=True): Runs a clustering algorithm for a range of cluster numbers (2 to 10), computes silhouette scores and inertia, and prints the results.
  • plot_silhouette_analysis(X, labels, title): Plots the silhouette analysis for the clustering results.
  • plot_heatmap(Y, cmap=[green, red, blue, yellow]): Plots a heatmap of the given matrix Y.
  • visualize_clusters(X, labels, title): Visualizes clusters by plotting time series data that belong to the same cluster.

Jupyter Notebook

The ProjetUL.ipynb notebook demonstrates the use of the utility functions on a dataset. It includes the following steps:

  1. Read the data: Load the dataset from the provided URLs.
  2. Calculate the distance matrix: Compute the distance matrix using Euclidean and DTW distances.
  3. Clustering: Apply KMedoids and KMeans clustering methods.
  4. Visualization: Use elbow method, silhouette scores, heatmaps, and cluster visualizations to analyze the results.

Example Code

# Import utility functions
from utils import *

# Load the data
X = pd.read_csv("http://allousame.free.fr/mlds/donnees/X.txt", sep=" ", header=None)

# Calculate distance matrix using Euclidean distance
from scipy.spatial.distance import pdist, squareform
dist_matrix_euc = pdist(X, metric='euclidean')
dist_matrix_euc = squareform(dist_matrix_euc)

# Run KMeans clustering
from sklearn.cluster import KMeans
clust_method = lambda k: KMeans(n_clusters=k, random_state=0, n_init='auto')
inertias, silhouettes, labels = run_clustering_method(clust_method, X, metric="euclidean")

# Plot results
plot_inertia(inertias, title="Elbow method for KMeans with Euclidean distance")
plot_silhouette(silhouettes, title="Silhouette score for KMeans with Euclidean distance")
visualize_clusters(X, labels, title="Visualizing KMeans clusters with Euclidean distance")

Conclusion

This project provides a comprehensive example of clustering time series data and visualizing the results using various techniques. The provided utility functions and Jupyter notebook can be adapted to different datasets and clustering methods.