Spatiotemporal Traffic Forecasting in Smart Cities: Evaluating Deep Learning Approaches with IoT-Driven Data

Abstract

The population of cities is increasing rapidly, and the numerous usage of on-demand mobility services has subjected the current transportation infrastructure to unprecedented demands, which makes precise traffic flow prediction an urgent need to Intelligent Transportation Systems (ITS). The nonlinear and dynamic, even spatiotemporal characteristics of city traffic are becoming increasingly ineffective to be modeled using the traditional statistical and time-series models, including ARIMA and the Holt-Winters, especially when exogenous factors and the magnitude of the data flows produced by Internet-of-Things (IoT) sensors are in play.  In the research, we overcome these shortcomings by investigating the state-of-the-art deep-learning systems to predict traffic in cities. In particular, we create and train Long Short-Term Memory (LSTM) networks, Temporal Convolutional Networks (TCNs) and Transformer based models on large-scale real-world benchmark sets, such as METR-LA and PEMS-Bay.  Our preprocessing pipeline used was very strict to address the missing values, standardize the time-sequences and maintain chronological order.  The standard regression measures such as mean absolute error (MAE), root mean square error (RMSE) and mean absolute percentage error (MAPE) were used to assess model performance.  Through experimentation, it can be shown that Transformer-based architectures are consistently superior, and have better accuracy because they can better represent intricate global spatiotemporal interactions with self-attention functions. A trade-off exists between accuracy and computational efficiency in TCNs, but LSTMs can still be used in a case where a high level of scalability is not necessary. These results support the effectiveness of spatiotemporal modelling based on deep learning in predicting traffic and emphasize its possible potential to support proactive control of traffic, minimize congestion, decrease fuel use and emissions, and improve the overall sustainability and safety of urban transportation..