Creating Features from a Time Series with tsfresh

Throughout this book, we’ve discussed feature engineering methods and tools suitable for tabular data and relational datasets. In this chapter, we will focus on time series data. A time series is a sequence of observations taken sequentially over time. Examples of time series are energy generation and demand, temperature, air pollutant concentration, stock prices, and sales revenue. Each of these examples constitutes a variable, and their values change over time.

The availability of cheap sensors that can measure motion, movement, humidity, or temperature, among several other things, has dramatically increased the availability of temporally annotated data. These time series can then be used in classification tasks. For example, based on the electricity fingerprint of a household at a given time interval, we can infer whether a certain appliance was being used. Based on the signal of an ultrasound sensor, we can determine the...