Abnormal Driving Detection using GPS Data.

Area of Science:

• Data Science
• Transportation Engineering
• Machine Learning

Background:

• Abnormal Driving Detection (ADD) is crucial for road safety and risk management.
• GPS datasets offer rich information for analyzing driving behaviors.
• Existing methods may require complex feature engineering or lack robustness.

Purpose of the Study:

• To develop and evaluate an integrated approach for Abnormal Driving Detection (ADD) using GPS data.
• To leverage dimensionality reduction and clustering for identifying anomalous driving patterns.
• To provide a robust method for driver safety and insurance risk assessment.

Main Methods:

• Data preprocessing and aggregation of GPS records (Speed Over Ground, Course Over Ground, longitude, latitude) into minute-level segments.
• Dimensionality reduction using Singular Value Decomposition (SVD).
• Clustering of driving patterns using the K-means algorithm.

Main Results:

• The integrated methodology effectively distinguishes between normal and abnormal driving behaviors.
• Singular Value Decomposition (SVD) successfully reduced data dimensionality.
• K-means clustering identified distinct driving patterns indicative of abnormal driving.

Conclusions:

• The proposed integrated approach offers a promising solution for Abnormal Driving Detection (ADD).
• This method has significant implications for improving driver safety and insurance risk assessment.
• Further research can explore personalized interventions based on detected driving patterns.