Understanding Time Series Clustering: Hands-On Hierarchical Clustering and Dynamic Time Warping (DTW)

Image created by the Author

Unlike supervised learning, which depends on predefined outcomes for model training, unsupervised learning facilitates the discovery of inherent similarities in the data without requiring predetermined results. This also holds promise in the context of time series.

However, when it comes to the execution of unsupervised classification algorithms on time series data, there are typically mounting questions regarding it´s correct implementation. For example, traditional clustering models, like K-Means, do not account for temporal dependencies and the sequential nature of time series. Characteristics, like trend, seasonality, and cyclicity, demand specialized approaches within the clustering process.

In this article, I aim to elaborate the process of time series clustering with the help of Dynamic Time Warping and Hierarchical Clustering. Furthermore, I will demonstrate its practical implementation based on an example of unsupervised stocks data classification.

So, let´s start our journey with some introductory chapters 📖!

What is time series data?

“Time series data is a collection of observations made sequentially over time.”