Meta-IP: An Imbalanced Processing Model Based on Meta-Learning for IT Project Extension Forecasts

Table 3

Comparison of AUC and BACC across models trained with various sampling methods on the IT project extension forecast tasks. Bold figures reflect the row maximum.

Model	1^st-year dataset		2^nd-year dataset		3^rd-year dataset		4^th-year dataset		5^th-year dataset
	AUC	BACC	AUC	BACC	AUC	BACC	AUC	BACC	AUC	BACC
Naive Bayesian	0.967	84.3 ± 0.11	0.958	83.2 ± 0.03	0.962	85.7 ± 0.10	0.74	87.2 ± 0.08	0.963	85.6 ± 0.09
Bagging	0.943	73.2 ± 0.05	0.957	77.2 ± 0.02	0.963	73.9 ± 0.04	0.962	78.3 ± 0.07	0.972	75.2 ± 0.09
SMOTE	0.952	70.8 ± 1.30	0.952	66.8 ± 0.90	0.979	71.2 ± 1.20	0.955	69.3 ± 0.80	0.957	69.5 ± 1.10
SVM	0.921	79.2 ± 1.00	0.934	83.4 ± 0.90	0.958	81.3 ± 0.80	0.963	80.2 ± 1.10	0.954	82.7 ± 1.20
SMEOTE+	0.955	72.4 ± 0.02	0.945	74.3 ± 0.03	0.935	72.7 ± 0.01	0.937	77.3 ± 0.05	0.943	74.8 ± 0.07
SMOTE+	0.946	73.5 ± 0.03	0.927	72.7 ± 0.01	0.947	73.5 ± 0.04	0.953	76.7 ± 0.03	0.938	72.8 ± 0.05
Meta-IP	0.975	89.7 ± 0.05	0.965	90.1 ± 0.06	0.988	91.2 ± 0.02	0.985	89.3 ± 0.03	0.976	91.8 ± 0.07