Dynamic Road Anomaly Detection: Harnessing Smartphone Accelerometer Data with Incremental Concept Drift Detection and...

Area of Science:

• Transportation engineering
• Data science
• Machine learning

Background:

• Effective road condition monitoring is vital for safe transportation.
• Current methods often rely on pre-trained models, limiting adaptability.
• Real-time, incremental detection is needed for dynamic environments.

Purpose of the Study:

• To develop a novel, real-time road condition monitoring technique.
• To address limitations of pre-trained models in dynamic environments.
• To improve the accuracy and robustness of anomaly classification.

Main Methods:

• Utilized crowd-sourced smartphone sensor data.
• Implemented a hybrid anomaly detection method (unsupervised and supervised learning).
• Focused on incremental learning to manage concept drift.

Main Results:

• Achieved a 96% success rate in road anomaly detection.
• Demonstrated significant improvements in accuracy and robustness.
• Showcased effective management of concept drift.

Conclusions:

• Incremental learning enhances model responsiveness for road anomaly detection.
• The proposed hybrid method offers a promising direction for transportation safety.
• This technique supports future resource optimization strategies.