Statistical models have their limitations, especially when a dataset is large and has many features and nonlinear relationships. In such cases, deep learning is the perfect tool for time series forecasting. In this part of the book, we’ll work with a massive dataset and apply different deep learning architectures, such as long short-term memory (LSTM), a convolutional neural network (CNN), and an autoregressive deep neural network, to predict the future of our series. Again, we’ll conclude this part with a capstone project to test your skills.
Deep learning is a subset of machine learning, and it is therefore possible to use more traditional machine learning algorithms for time series forecasting, such as gradient-boosted trees. To keep this section reasonable, we won’t cover those techniques specifically, although data windowing is required to forecast time series with machine learning, and we’ll apply this concept numerous times.