Summary

In this chapter, we have described three deep learning-based approaches to develop time series forecasting models. Neural networks are suitable in cases where there is little information about the underlying properties such as long-term trend and seasonality or these are too complex to be modeled with an acceptable degree of accuracy by traditional statistical methods. Different neural network architectures such as MLP, RNN, and CNN extract complex patterns from the data. If neural network models are trained with appropriate measures to avoid overfitting on training data, then these models generalize well on unseen validation or test data. To avoid overfitting, we applied dropout, which is widely used in deep neural networks for a variety of datasets and applications. We hope that this chapter gives you an idea of advanced techniques available for time series forecasting. The Jupyter notebooks accompanying the chapter are expected to give you the necessary base knowledge, which would...