Table 4 Evaluation of experiments to investigate drifts between internal and external data.

From: Studying and mitigating the effects of data drifts on ML model performance at the example of chemical toxicity data

 

Liver toxicity

Micro nucleus test

CV

Predict holdout set

CV

Predict holdout set

Cal_original

Cal_update

Cal_original

Cal_update

Balanced validity

0.81

0.47

0.82

0.82

0.50

0.74

Balanced efficiency

0.81

0.89

0.38

0.79

0.94

0.40

Balanced accuracy

0.77

0.43

0.49

0.77

0.49

0.39

Validity inactive class

0.81

0.75

0.84

0.80

0.99

0.61

Efficiency inactive class

0.84

0.84

0.45

0.79

0.89

0.54

Accuracy inactive class

0.77

0.70

0.63

0.75

0.99

0.29

Validity active class

0.82

0.20

0.80

0.83

0.00

0.88

Efficiency active class

0.78

0.95

0.31

0.79

1.00

0.26

Accuracy active class

0.77

0.16

0.35

0.78

0.00

0.50

Validity

0.82

0.58

0.84

0.81

0.66

0.70

Efficiency

0.80

0.87

0.40

0.79

0.93

0.45

Accuracy

0.77

0.52

0.57

0.76

0.63

0.33

Back to article page