Anomaly Detection Using Multiple Data-based iForest for Realizing Digital Twin in Winter Road

抄録

Winter road environments pose significant challenges to transportation safety and maintenance due to adverse weather conditions. This study proposes an anomaly detection model that leverages multiple databased Isolation Forest(iForest) to assess road conditions. This study integrates the data from an on-board edge system with precipitation data from XRAIN. The features related to winter road were selected, and labels were constructed for training iForest, a tree-based unsupervised learning method. This study proposes training the model only on readily obtainable and verifiable normal data and evaluates whether its performance can approximate that of supervised models requiring perfectly labeled datasets. Experiment results demonstrate the proposed model’s effectiveness in detecting anomalies under winter road conditions. Compared with other unsupervised learning techniques, the iForest model achieved the highest performance. Although supervised learning models output higher performance, their reliance on perfectly labeled data, which is difficult to acquire, limits their practicality in this context. In contrast, the results of our model are relatively closer to best performance, so we adopted it.

The findings highlight the practical significance of the proposed method for road monitoring and maintenance, providing a robust, low-cost solution for anomaly detection in complex winter road environments. This research not only enhances decision-making for traffic safety and resource allocation but also contributes to advancing the development of digital twin systems for intelligent transportation management.