Table 6 Results of Kruskal–Wallis H-tests for each dataset. For a fixed dataset d, the null hypothesis is “the median NS-AUC value \(\mu _\text {NN}(d)\) obtained on dataset d by pair-oriented neural networks is equal to the median NS-AUC value \(\mu _\text {GB}(d)\) on the same dataset by pair-oriented graph-based approaches”. In each test, the number of elements in each group is \(N=200\). The values are rounded up to the closest first or second decimal places. All tests on adjusted p-values are significant at level \(\alpha =1\%.\).

From: Comprehensive evaluation of pure and hybrid collaborative filtering in drug repurposing

Dataset

Cdataset

LRSSL

PREDICT

DNdataset

TRANSCRIPT

Fdataset

PREDICT (public)

Gottlieb

H

26.4

43.0

70.8

84.1

97.3

128.5

143.2

144.1

\(\mu _{NN}-\mu _{GB}\)

\(-0.07\)

\(-0.06\)

\(-0.09\)

\(-0.21\)

\(-0.08\)

\(-0.11\)

\(-0.11\)

\(-0.11\)