TABLE 4.

Model evaluation in multi-label ensembles using LR and XGBoost correspondingly.

Criteria\Ensembles	BinaryRelevance LR (XGBoost)	ClassifierChain LR (XGBoost)	Lowest Power set LR (XGBoost)
Hamming loss	0.2023 (0.1622)	0.2015 (0.1628)	0.2044 (0.1524)
Accuracy score	0.6863 (0.7334)	0.6885 (0.7553)	0.7119 (0.7717)
Jaccard score	0.6019 (0.6677)	0.6038 (0.6676)	0.6033 (0.6839)

Hamming loss, accuracy score, and Jaccard score are used to evaluate the multi-label model primarily. Hamming loss refers to the average fraction of the wrong prediction of each sublabel. The accuracy score is based on the accuracy of the serial label prediction. Jaccard score measures the proportion of prediction for a sample to its true label. Bold values refer to better performance based on each criterion.