Forecasting Time Series Data using Recurrent Neural Networks: A Systematic Review
Manpreet Kaur Bhatia, Vinayak Bhatt · Journal for Research in Applied Sciences and Biotechnology · 2024
The method of time series forecasting stands crucial in multiple application areas that include finance as well as healthcare and energy management and climate modeling. RNNs serve as a powerful tool under deep learning because they possess ability to detect sequential data patterns while extracting temporal dependencies from time series data using traditional statistical methods which were previously the dominant approach. This paper conducts an organized review of modern techniques for predicting time series data by using RNNs. This discussion covers three major RNN architectures together with Long Short-Term Memory (LSTM) networks and Gated Recurrent Units (GRUs) as well as their combination with hybrid models. The paper examines how RNN-based models perform against traditional approaches before addressing RNN-based forecasting problems and suggesting potential research paths for the future. The analysis reviews multiple performance indicators utilized in past research to establish profound knowledge about RNN-based forecasting methods. The paper examines RNN benefits while analyzing the computational limitations and overfitting risks and interpretability problems that RNN systems encounter. The review investigates new frameworks including attention systems together with strengthening strategies and combination methods of statistical analysis with machine learning structures. Research outcomes demonstrate that RNN models particularly LSTM and GRU achieve great forecasting precision but future application research needs to optimize execution performance and advance interpretability capabilities of these models.