Classification of freeway traffic patterns for incident detection using constructive probabilistic neural networks.

Area of Science:

• Artificial Intelligence
• Transportation Engineering
• Machine Learning

Background:

• Freeway incident detection is crucial for traffic management and safety.
• Existing methods often struggle with adaptability to diverse traffic conditions.
• The need for efficient and accurate real-time incident detection systems is paramount.

Purpose of the Study:

• To introduce a novel constructive probabilistic neural network (CPNN) for enhanced freeway incident detection.
• To evaluate the performance and adaptability of the CPNN model across different freeway environments.
• To demonstrate the model's efficiency in terms of detection rates, false alarm rates, and computational size.

Main Methods:

• Development of a CPNN integrating clustering and automated training for incident detection.
• Model training and validation on the Ayer Rajah Expressway (AYE) in Singapore.
• Model adaptation and testing on the I-880 freeway in California.

Main Results:

• Achieved 92% detection rate and 0.81% false alarm rate on AYE.
• Attained 91.30% detection rate and 0.27% false alarm rate on I-880 after adaptation.
• Network pruning reduced model size by a factor of 11, and adaptation further reduced it by 50 times.

Conclusions:

• CPNN demonstrates superior performance and adaptability for freeway incident detection.
• The proposed method offers significant model size reduction compared to conventional probabilistic neural networks.
• CPNN is a promising adaptive classifier for dynamic traffic environments.