Fig. 3
From: Comprehensive evaluation of pure and hybrid collaborative filtering in drug repurposing
Boxplots of testing metric values for the Top-3 algorithms (in average) across \(N=100\) iterations for each dataset in Table 1, for a specific training/testing set splitting method. PREDICT(p) corresponds to the public version of PREDICT, whereas PREDICT refers to the private version of the dataset. (a) AUC values for randomly split sets. (b) AUC values for weakly correlated sets. (c) NS-AUC values for randomly split sets. (d) NS-AUC values for weakly correlated sets.
