Table 1 Machine learning results in the transductive setting.

From: Mass spectra alignment using virtual lock-masses

Condition

AdaBoost

Decision Tree

SCM

L1-SVM

Clomiphene Detection

Binning only

98.0% (4.7)

98.6% (1.8)

95.2% (1.1)

89.6% (52.0)

VLM + Binning

98.2% (4.9)

97.0% (2.3)

97.0% (1.2)

93.6% (2.6)

VLM + Alignment

98.8% (2.3)

99.4% (1.0)

99.4% (1.0)

92.8% (138.6)

Acetaminophen Detection

Binning only

99.2% (1.0)

99.2% (1.0)

99.2% (1.2)

97.6% (97.5)

VLM + Binning

99.2% (1.0)

99.2% (1.0)

99.4% (1.0)

99.0% (121.0)

VLM + Alignment

99.8% (1.0)

100.0% (1.0)

99.4% (1.0)

99.6% (63.4)

Malaria Detection

Binning only

92.4% (51.8)

82.5% (4.3)

84.6% (2.2)

92.6% (150.1)

VLM + Binning

93.3% (39.7)

88.7% (4.6)

89.4% (2.0)

95.4% (133.2)

VLM + Alignment

93.8% (65.3)

86.1% (4.8)

85.4% (2.3)

95.2% (131.4)

Cancer Detection

Binning Only

70.4% (69.2)

63.8% (6.4)

55.6% (1.9)

56.8% (113.6)

VLM + Binning

70.2% (43.9)

61.6% (4.8)

53.6% (2.2)

69.4% (138.6)

VLM + Alignment

67.4% (30.0)

62.6% (2.3)

59.6% (2.2)

74.6% (135.2)

  1. The percentage in each column is the average accuracy of classifiers on 10 repeats of the experiment. The number shown in parentheses is the average number of features used by the classifiers. The algorithms tested were AdaBoost, the Decision Tree algorithm, the Set Covering Machine (SCM) and a L1-norm Support Vector Machine (L1-SVM).