The effect of sensor position deflection on behavior classification performance

The behavioral recognition system based on accelerometers can support the assessment of cow health. Machine learning algorithms can efficiently classify accelerometer data collected from cow-mounted sensors. However, with cow activities, the sensor may deviate from its original position, which may affect the accelerometer data collected, thereby affecting the performance of behavior classification. From the collected data, we generate deviated collar sensor simulation data to evaluate the classification performance of the model under different circumstances. In the case of using synchronized acceleration data from the leg and neck of cow, applying the Random Forest algorithm with mean and RMS features, the results showed that the behavioral classification performance did not change significantly when the collar-mounted sensor deviated.