Chapter 15: Advanced Techniques for Complex Time Series

Time series data can contain complex seasonality – for example, recorded hourly data can exhibit daily, weekly, and yearly seasonal patterns. With the rise of connected devices – for example, the Internet of Things (IoT) and sensors – data is being recorded more frequently. For example, if you examine classical time series datasets used in many research papers, many were smaller sets and recorded less frequently, such as annually or monthly. Such data contains one seasonal pattern. More recent datasets and research now use higher frequency data, recorded in hours or minutes.

Many of the algorithms we used in earlier chapters can work with seasonal time series. Still, they assume there is only one seasonal pattern, and their accuracy and results will suffer on more complex datasets.

In this chapter, you will explore new algorithms that can model a time series with multiple seasonality for forecasting...