Analysis of the paper
Li, H. (2019). Multivariate time series clustering based on common principal component analysis. Neurocomputing, 349, 239-247
In this project, we focus on a thorough analysis and independent implementation from scratch of the method. We evaluated this method on a new dataset, along with two of the datasets originally used by the authors. Our exploration also extended to examining the method’s limitations and potential for improvement through four specific experiments: assessing performance with unbalanced data, evaluating sensitivity to noise, determining the optimal number of principal components to use, and experimenting with different distance metrics for calculating the reconstruction error. Our report contains the paper analysis, our results and critical perspectives.
Files description:
Mc2PCA_class.py: Contains the implementation of the Mc2PCA algorithmload_data.py: Load the datasets from the raw fileseval.py: Contains the methods to compute evaluation metrics on the clustering resultsexperiments.ipynb: Contains all the experiments performed
- Python >= 3.8
To install the required dependencies, run the following command: pip install -r requirements.txt
Carnegie Mellon University Motion Capture Database. CMU Motion Capture Database S16 Source: https://timeseriesclassification.com/description.php?Dataset=Epilepsy
Kudo,Mineichi, Toyama,Jun, and Shimbo,Masaru. Japanese Vowels. UCI Machine Learning Repository. https://doi.org/10.24432/C5NS47. Loaded from the sktime library: https://www.sktime.net/en/latest/api_reference/auto_generated/sktime.datasets.load_japanese_vowels.html
Villar, J., Vergara, P., Men'endez, M., Cal, E., Gonz'alez, V., & Sedano, J. (2016). Generalized models for the classification of abnormal movements in daily life and its applicability to epilepsy convulsion recognition. International journal of neural systems, 26(06), 1650037. Source: https://timeseriesclassification.com/description.php?Dataset=Epilepsy