Traffic Volume Forecast Models with High Sensor Data Uncertainty

บทคัดย่อ

Traffic volume forecasting is an important task for the motorway planning and management. The performance of these forecasts is often degraded by the high uncertainty of sensor data, particularly when the data are subject to delay. This study aims to develop methods for imputing and forecasting traffic volume under high uncertainty and delayed data conditions. The objective is to enhance the precision of predictions for traffic volume. This study introduced a new data imputation method as well as a sequence-based machine learning model, namely, Long Short-term Memory (LSTM) model, to handle highly uncertain sensor data. The model's performance is evaluated using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE) with a result of 26.67 vehicles, 17.31 vehicles and 9.26% respectively. Specifically, the model demonstrated a high level of sensitivity to delayed data, 17.45% of delayed data, meaning that it was able to accurately adjust its predictions based on changes in data availability and processing times. This suggests that the proposed approach has the potential to significantly improve the accuracy and reliability of traffic volume forecasting in real-world settings, where delays and disruptions are common occurrences. Overall, our study provides strong evidence for the efficacy of the proposed approach in the face of delayed data and highlights its potential as a valuable tool for traffic management and planning in Thailand and beyond.