On Multilabel Classification Methods of Incompletely Labeled Biomedical Text Data

Table 11

Microaveraged results for RF, trained on “incompletely labeled” Yeast dataset (with different fraction of deleted labels 𝑝) with add+WkNN label restoration. Optimal WkNN parameters 𝑘 and 𝑇 are acquired via grid search.

𝑝 parameter	Optimal 𝑘	Optimal 𝑇	Precision	Recall	𝐹1
0	—	—	—	—	—
0.1	10	0.3	0.5940	0.7120	0.6224
0.2	10	0.25	0.5626	0.7462	0.6146
0.3	10	0.15	0.5282	0.7992	0.6095
0.4	10	0.10	0.5223	0.7648	0.5940
0.5	10	0.05	0.4672	0.8388	0.5726
0.6	15	0.05	0.4547	0.8695	0.5707