Time Series K-Means Clustering with Autoencoder Feature Extraction

This is a class project for Data Science and Machine Learning course of Università di Salerno Computer Science Master degree. In this repo there are two Jupyter Nootebook in which are experimented two clustering approaches for time series. One relies on an autoencorder to extract TS features an then cluster them with K-Means. The other uses TSLearn for DTW based K-Means.

Installation

Prerequisites

Python 3 is required. If you want to run TSLearn notebook, also C++ Build Tools are needed.

Steps

1. Download the repo;
2. Go in the repo source folder
3. (Optional) Install a virtual envirorment
4. Run pip install -r requirements.txt
5. Run jupyter notebook to open Jupyter
6. Notebooks are located in Run jupyter/ folder

Used Datasets

Time series are not multivariate

• fordA
• fordB
• ECG5000
• ECG200
• phalangesOutlinesCorrect
• TwoPatterns
• ChorelineConcetration
• refrigerationDevice
• TwoLeadECG