Short-Term Traffic Prediction Using Deep Learning Long Short-Term Memory: Taxonomy, Applications, Challenges, and Future Trends

Abstract

This paper surveys the short-term road traffic forecast algorithms based on the long-short term memory (LSTM) model of deep learning. The algorithms developed in the last three years are studied and analyzed. This provides an in-depth and thorough description of the algorithms rather than their marginal description as performed in the existing surveys that focus on general deep learning algorithms. The chosen algorithms are classified depending upon the use of LSTM in combination with other techniques for processing input data features towards a final traffic forecast. The operational strategies of the algorithms are described with merits and limitations. Moreover, a comparative analysis of the compared classes of algorithms is also provided. These strategies are helpful in selection of the right algorithms and their classes for the diverse traffic conditions and their future investigation for improvement. Besides, the applications of these classes of algorithms to traffic forecast in various networks for the latest decade is graphically depicted. Moreover, the applications of the LSTM in other fields involving a forecast are provided. Finally, the challenges associated with the short-term traffic forecast using the LSTM are described and strategies are highlighted for their future investigation.INDEX TERMS short-term traffic prediction, long short-term memory, LSTM, deep learning, intelligent transportation.

Dynamic Telemetry and Deep Neural Networks for Anomaly Detection in 6G Software-Defined Networks

“…Artificial neural network models have been used for several decades, but a lack of computational resources have led to the dominance of other machine learning techniques. Nowadays, the trend associated with the type of model strongly focuses on deep learning [21,22].…”

“…In this article, a specific type of neural network called long short-term memory (LSTM) is utilized. It is employed for predicting time series and exhibits excellent characteristics in capturing periodic dependencies [21,22].…”