SIMPD: an algorithm for generating simulated time splits for validating machine learning approaches

Table 1 Summary of the distribution changes for the descriptors chosen

Property	Sign	Frac	Median (frac change)	Median (train)	Number of projects
SA score	1	0.88	0.09	2.8	109
HeavyAtomCount	1	0.75	0.09	31.0	114
TPSA	1	0.76	0.14	88.6	109
fr_benzene/1000 HeavyAtoms	-1	0.81	0.19	43.5	118

The meanings of the columns are: Sign = sign of the difference between training and test; Frac = fraction of the data sets showing a change with that sign; Median(frac change) = median fractional change of the value across the data sets; Median(train) = median value of the property in the training set; Number of projects = number of projects where the difference in the training/test distributions was statistically significant (see text)

ISSN: 1758-2946