Time Series Forecasting: A Comparison of Deep Neural Network Techniques

Authors

  • Banalaxmi Brahma*, Rajesh Wadhvani

Abstract

Time series forecasting is an invaluable task for many problems across a wide range of
domains, including prediction of air quality, solar irradiance, electricity consumption, stock prices and
so on. Traditionally the statistical and autoregressive models were used for the majority part of
forecast applications. However, the real world applications involve data with complex and non-linear
dependencies consisting of long-term as well as short-term patterns, for which traditional approaches
might fail. The Recurrent Neural Networks were introduced to model such sequential data and here we
aim to review the variants of Recurrent Neural Networks along with a special class of sequential
modeling known as attention mechanism. Here we review and analyze the deep neural network based
models focusing mainly on the task of time series forecasting utilizing recurrent neural networks

Published

2020-06-30

Issue

Section

Articles