Table 4.

Clinical trial termination classification results, using single model without random under sampling (a), and random under sampling based ensemble model (b) trained using all features. A * indicates where the ensemble classifier is significantly different from its single model classifier counterpart at $p<0.05$, and ** indicates a higher level of confidence at $p<0.001$.

(a) Single model termination classification
Model	Accuracy	Balanced	F1-Score	AUC
Neural Network	88.47%	50.23%	1.21%	71.71%
Random Forest	88.54%	50.02%	0.10%	71.67%
XGBoost	88.55%	50.26%	1.18%	72.81%
Logistic Reg.	88.46%	50.48%	2.34%	71.42%

(b) Ensemble model termination classification
Model	Accuracy	Balanced	F1-Score	AUC
Neural Network	62.66%**	66.42%**	30.43%**	72.03%
Random Forest	66.33%**	66.58%**	31.28%**	72.59%*
XGBoost	63.92%**	67.20%**	31.21%**	73.01%
Logistic Reg.	63.31%**	65.79%**	30.11%**	71.46%