Detecting the Risk of Customer Churn in Telecom Sector: A Comparative Study

Table 6

Performance of RF on different undersampling strategies.

Dataset	Optimization algorithm	Accuracy	Recall	Precision	AUC	F1-score	MAE
Dataset 1	0.3 Undersampling
	GS	0.99	1.00	0.91	0.99	0.97	0.014
	RS	0.94	0.74	0.83	0.85	0.88	0.059
	GA-DEAP	0.93	0.68	0.82	0.82	0.85	0.068
	GA-TPOT	0.96	0.87	0.84	0.91	0.91	0.043
	0.6 Undersampling
	GS	0.96	0.96	0.79	0.95	0.92	0.043
	RS	0.92	0.89	0.67	0.90	0.86	0.079
	GA-DEAP	0.93	0.88	0.71	0.91	0.87	0.074
	GA-TPOT	0.92	0.88	0.67	0.90	0.85	0.082
Dataset 2	0.4 Undersampling
	GS	0.95	0.93	0.90	0.95	0.94	0.048
	RS	0.83	0.61	0.73	0.76	0.77	0.174
	GA-DEAP	0.83	0.63	0.72	0.76	0.78	0.173
	GA-TPOT	0.83	0.65	0.73	0.77	0.79	0.167
	0.7 Undersampling
	GS	0.89	0.96	0.73	0.91	0.88	0.108
	RS	0.78	0.73	0.59	0.76	0.75	0.218
	GA-DEAP	0.80	0.78	0.62	0.79	0.77	0.196
	GA-DEAP	0.80	0.78	0.62	0.79	0.77	0.196
	GA-TPOT	0.80	0.75	0.62	0.78	0.77	0.201