Research Article
Detecting the Risk of Customer Churn in Telecom Sector: A Comparative Study
Table 4
Performance evaluation of applying HPO methods to the SVM classifier based on low-ratio undersampling strategy (0.3 and 0.4).
| Dataset | Optimization algorithm | Accuracy | Recall | Precision | AUC | F1-score | MAE |
| Dataset 1 | Default HPs | 0.89 | 0.36 | 0.73 | 0.66 | 0.71 | 0.113 | GS | 0.92 | 0.65 | 0.76 | 0.80 | 0.83 | 0.081 | RS | 0.92 | 0.67 | 0.75 | 0.81 | 0.83 | 0.080 | GA-DEAP | 0.92 | 0.66 | 0.74 | 0.81 | 0.83 | 0.083 | GA-TPOT | 0.92 | 0.65 | 0.75 | 0.80 | 0.83 | 0.082 |
| Dataset 2 | Default HPs | 0.77 | 0.46 | 0.64 | 0.67 | 0.69 | 0.226 | GS | 0.78 | 0.54 | 0.64 | 0.71 | 0.72 | 0.215 | RS | 0.78 | 0.54 | 0.64 | 0.71 | 0.72 | 0.215 | GA-DEAP | 0.83 | 0.63 | 0.72 | 0.76 | 0.78 | 0.173 | GA-TPOT | 0.77 | 0.50 | 0.62 | 0.68 | 0.70 | 0.229 |
|
|