Skip to main content

Table 8 Accuracy of TL for predicting the fourth level ATC code

From: Predicting anatomic therapeutic chemical classification codes using tiered learning

4th level ATC Code Prediction using different initializations and comparison to STP

Initialization method

Supervised learning

SuperPred

ChEMBL (Perfect 1st level)

STP

Prediction method

A(4)

CA(4)

(N(4))

A(4)

CA(4)

(N(4))

A(4)

CA(4)

(N(4))

A(4)

CA(4)

(N(4))

MLP

48.57%

63.75%

(37.81%)

48.57%

65.14%

(37.09%)

50.90%

75.58%

(41.93%)

51.61%

67.97%

(40.32)

Naive Bayes

36.92%

47.45%

(19.35%)

45.70%

62.59%

(29.03%)

46.77%

70.30%

(32.80%)

38.35%

55.20%

(22.76%)

Random Forest

40.68%

57.12%

(29.93%)

47.49%

63.66%

(33.87%)

50.36%

74.24%

(38.53%)

39.43%

61.20%

(28.49%)

SVM

50.18%

63.53%

(40.68%)

56.09%

68.50%

(45.34%)

56.81%

78.90%

(49.64%)

41.22%

65.37%

(41.22%)

  1. The best STP results are included for comparison. The highest prediction accuracies obtained with each initialization strategy are highlighted in bold