Feature Selection Model Based on IWOA for Behavior Identification of Chicken

Area of Science:

• Machine Learning
• Animal Behavior Analysis
• Sensor Data Processing

Background:

• Redundant features in accelerometer data can negatively impact behavior recognition model performance.
• Improving recognition accuracy is crucial for effective animal behavior monitoring.

Purpose of the Study:

• To propose an improved Whale Optimization algorithm (IWOA) combined with extreme gradient boosting (XGBoost) for chicken behavior identification.
• To reduce feature dimensions and enhance the accuracy of accelerometer-based behavior recognition.

Main Methods:

• Utilized a nine-axis inertial sensor to collect chicken behavior data.
• Applied noise reduction and sliding window techniques to extract 44 time and frequency domain features.
• Implemented IWOA with a mixed strategy, incorporating good point set, adaptive weight, and dimension-by-dimension lens imaging learning for feature selection.

Main Results:

• Reduced feature dimensions by 72.73%.
• Increased behavior recognition accuracy to 95.58% (a 2.41% improvement).
• The IWOA-XGBoost model demonstrated superior accuracy compared to other dimensionality reduction methods.

Conclusions:

• The proposed IWOA-XGBoost method effectively reduces feature dimensions and improves chicken behavior recognition accuracy.
• The dimension reduction results show universality across different classification algorithms.
• This approach offers a robust method for behavior recognition using acceleration sensor data.