A Conceptual CNN–LSTM Framework for Acoustic Traffic Congestion Forecasting on Low-Cost Edge Computing Platforms

Abstract

Traffic congestion in urban areas is still one of the key obstacles of intelligent transportation systems at present, which has implications for mobility, energy consumption, and environmental sustainability. Current methods in traffic monitoring have been based on infrastructure-intensive sensing technologies such as cameras, inductive loop detectors, and GPS-based methods, which are generally expensive and sensitive to environmental conditions. In the meantime, acoustic traffic monitoring has recently proved to be a cost-effective and robust method by making use of traffic-generated sound patterns to infer traffic conditions. Some previous research showed that deep learning models, especially Long Short-Term Memory (LSTM) networks, could produce high classification accuracy in acoustic traffic analysis with real-world datasets. But most existing approaches are only directed towards recognition of traffic states without considering the need to predict traffic congestion forecasting. In this paper, we implement a conceptual hybrid CNN–LSTM for acoustic traffic congestion forecasting on low-cost edge computing platforms for analysis. Furthermore, the framework extends previous classification-based work by rendering the question as a time-series forecasting problem. Both Convolutional Neural Networks (CNNs) are used to extract spectral representations from acoustic feature maps together with LSTM networks to capture temporal dependencies in sequential traffic patterns. This unified architecture promotes prediction-based traffic analytics while offering compatibility for resource-constrained edge devices. The proposed framework is constructed on top of pre-validated acoustic traffic datasets and experimental findings, thus lays the methodological basis for further forecasting-centered applications. It is accepted that the current study lacks experimental validation, and a complete examination, including comparison with baseline models using forecasting metrics, such as MAE and RMSE, is suggested as future work. In summary, the presented CNN–LSTM framework provides a structured basis to advance acoustic traffic monitoring beyond classification to forecasting-based intelligent transportation applications.