Skip to main content

Table 3 Accuracy of classifiers at levels one through four of the ATC code in the STP setting

From: Predicting anatomic therapeutic chemical classification codes using tiered learning

STP prediction accuracy at each level of the ATC code

ATC Level

1

2

3

4

Descriptor

CIP

CFP

CIP

CFP

CIP

CFP

CIP

CFP

MLP

73.66%

53.05%

67.56%

43.37%

59.14%

41.22%

33.87%

34.05%

Naive Bayes

47.67%

31.18%

36.38%

17.03%

30.47%

17.20%

22.76%

12.19%

Random Forest

66.49%

49.28%

58.96%

40.32%

50.18%

37.46%

28.49%

32.26%

SVM

72.22%

43.19%

64.52%

42.29%

58.60%

43.19%

41.22%

38.35%

  1. The classifiers were trained on 1674 compounds in the training set and tested on the 558 compounds using the CIP and CFP descriptors. The highest prediction accuracy at each ATC code level is highlighted in bold