Neural Networks, Deep Learning, and Tree‐Based Methods
Ruey S. Tay, Rong Chen · Wiley series in probability and statistics · 2018
This chapter introduces some recent developments in high-dimensional statistical analysis that are useful in time series analysis. The methods include neural networks, deep learning, and regression trees. The chapter focuses on the analysis of dynamically dependent data and their applications. Deep learning and neural networks have a long history with many successful applications and continue to attract lots of research interest. Neural networks are semi-parametric statistical procedures that have evolved over time with the advancement in computational power and algorithms. They can be used in prediction or in classification. The chapter describes the widely used vanilla feedforward networks. The concept of neural networks originates from models for the human brain, but it has evolved into a powerful statistical method for information processing. Application of neural networks often divides the data into training and forecasting subsamples. The chapter considers the deep belief nets (DBN) to gain further insight concerning deep learning.