A State-of-the-Art Review of Time Series Forecasting Using Deep Learning Approaches

Abstract

Time series forecasting has recently emerged as a crucial study area with a wide spectrum of real-world applications. The complexity of data processing originates from the amount of data processed in the digital world. Despite a long history of successful time-series research using classic statistical methodologies, there are some limits in dealing with an enormous amount of data and non-linearity. Deep learning techniques effectually handle the complicated nature of time series data. The effective analysis of deep learning approaches like Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Long short-term memory (LSTM), Gated Recurrent Unit (GRU), Autoencoders, and other techniques like attention mechanism, transfer learning, and dimensionality reduction are discussed with their merits and limitations. The performance evaluation metrics used to validate the model's accuracy are discussed. This paper reviews various time series applications using deep learning approaches with their benefits, challenges, and opportunities.