An Expressway ETC Missing Data Restoration Model Considering Multi-Attribute Features.

Area of Science:

• Intelligent Transportation Systems
• Data Mining
• Machine Learning

Background:

• Electronic toll collection (ETC) data mining is crucial for intelligent expressways.
• Ensuring ETC data integrity is vital for data quality.
• Deep learning applications for structured data restoration, like ETC data, are emerging.

Purpose of the Study:

• To propose a novel model for restoring missing expressway ETC transaction data.
• To address limitations in current deep learning applications for structured data restoration.
• To improve the accuracy and stability of ETC data restoration for big data analysis.

Main Methods:

• Utilized an entity embedding neural network (EENN) for categorical feature representation.
• Employed long short-term memory (LSTM) neural networks to capture vehicle speed patterns.
• Integrated processed features with MLP for comprehensive ETC data restoration.

Main Results:

• The proposed multi-attribute feature (MAF) model significantly outperformed existing methods in restoration accuracy on real ETC datasets.
• Achieved a 19.06% improvement in restoration accuracy on non-holiday datasets, with optimal MAE (12.394) and RMSE (23.815).
• Demonstrated a 5.82% increase in restoration stability on holiday datasets, with EENN and LSTM contributing significantly to accuracy and stability.

Conclusions:

• The proposed MAF model effectively enhances ETC data quality.
• The method meets the timeliness requirements for big data mining analysis in intelligent transportation.
• This research advances the application of deep learning in structured data restoration for intelligent expressways.