Short-term traffic flow prediction based on combination of SSA and LSTM

Short-term traffic flow prediction stands as a critical and highly topical research subject within the intelligent transportation system (ITS) domain. Owing to its inherent uncertainty and complexity, the task of predicting short-term traffic flow consistently presents a formidable challenge. Data driven methods have been achieved a success in the traffic prediction filed. In this paper, to enhance the precision of short-term traffic flow forecasting, we propose a hybrid model that integrates Singular Spectrum Analysis (SSA) with Long Short-Term Memory (LSTM) networks, denoted as SSA-LSTM. Initially, the traffic flow time series data is decomposed by SSA. Subsequently, the reconstructed traffic flow data is used to train the LSTM model. The model's efficacy is then validated through case analysis using real data collected from an expressway station in Jinan, Shandong, China. Finally, the SSA-LSTM model is compared with several models commonly used in short-term traffic flow prediction, including SSA and K-nearest neighbor (SSA-KNN), SSA and support vector regression (SSA-SVR), as well as single LSTM, KNN, and SVR models. The results of error analysis show that the proposed model (SSA-LSTM) has the best performance.