Skip to main content

Advertisement

Table 2 Performance comparison of four different machine-learning algorithms using the same input data set

From: Classification of behaviour in housed dairy cows using an accelerometer-based activity monitoring system

Behaviour Performance metric Classifier algorithm
1-min window Decision-tree K-means HMM SVM
Lying Sensitivity 74.09 85.93 90.17 92.91
  Precision 96.57 91.88 85.41 89.65
Standing Sensitivity 82.08 59.50 38.35 51.65
  Precision 47.01 29.28 37.28 77.01
Feeding Sensitivity 95.65 59.92 83.83 98.01
  Precision 92.03 86.13 91.54 91.01
Overall Sensitivity 83.94 68.45 70.78 80.85
  Precision 78.53 69.09 71.41 85.89
5-min window     
Lying Sensitivity 74.09 55.37 80.31 92.91
  Precision 97.95 98.10 94.51 91.66
Standing Sensitivity 88.46 69.23 76.92 60.89
  Precision 47.92 69.23 54.05 79.15
Feeding Sensitivity 97.44 99.36 97.44 98.29
  Precision 93.25 87.08 93.25 92.36
Overall Sensitivity 86.66 74.65 84.89 84.03
  Precision 79.71 84.80 80.60 87.72
10-min window     
Lying Sensitivity 77.42 80.65 70.97 89.60
  Precision 98.63 96.15 100.00 93.35
Standing Sensitivity 88.00 76.00 92.00 68.00
  Precision 55.00 59.38 50 76.04
Feeding Sensitivity 98.78 98.78 100 100.00
  Precision 93.10 90.00 93.18 93.18
Overall Sensitivity 88.06 85.14 87.65 85.86
  Precision 82.24 81.84 81.06 87.52
  1. Performance measures (sensitivity and precision) were obtained using 1-min, 5-min and 10-min windows. HMM refers to the hidden Markov model, and SVM refers to the support vector machine algorithm. Overall sensitivity is calculated as the arithmetic mean sensitivity for the three behaviours. Overall precision is calculated in a similar manner
  2. Values marked in bold indicate the best performing algorithm for each behaviour classification