Vehicle Classification Using an Imbalanced Dataset Based on a Single Magnetic Sensor.

Area of Science:

• Engineering
• Computer Science
• Data Science

Background:

• Automatic vehicle classification is crucial for intelligent transportation systems.
• Imbalanced datasets pose a significant challenge to classifier accuracy.
• Existing methods struggle with uneven class distributions in vehicle magnetic data.

Purpose of the Study:

• To develop an accurate automatic vehicle classifier for imbalanced datasets.
• To propose a novel feature extraction method for vehicle classification using magnetic sensor data.
• To evaluate the effectiveness of the Synthetic Minority Over-sampling Technique (SMOTE) in improving classifier performance.

Main Methods:

• Utilized a single anisotropic magnetoresistive sensor for data collection.
• Extracted features from both time and frequency domains.
• Employed k-nearest neighbor (KNN), support vector machine (SVM), and back-propagation neural network (BPNN) algorithms.
• Applied the SMOTE algorithm to address dataset imbalance.

Main Results:

• Achieved a classification accuracy of 95.46% using the KNN classifier combined with SMOTE.
• Demonstrated significant improvement in classification performance on imbalanced vehicle magnetic datasets.
• The proposed feature extraction method effectively captures relevant vehicle characteristics.

Conclusions:

• The combination of a novel feature extraction method and the SMOTE algorithm significantly enhances vehicle classification accuracy for imbalanced datasets.
• The KNN classifier, augmented with SMOTE, proves highly effective in overcoming data imbalance issues.
• This approach offers a robust solution for real-world automatic vehicle classification challenges.