Research Article

Negative Correlation Learning for Customer Churn Prediction: A Comparison Study

Table 6

Evaluation results (results of best five models are shown in bold).

Model Accuracy Actual churn rate Hit rate

-Nearest neighbour (IBK) 0.927 0.022 0.067
Naive Bayes (NB) 0.597 0.901 0.115
Random Forest (RF) 0.940 0.006 0.109
Genetic programing (GP) 0.759 0.638 0.142
Single ANN with BP 0.941 0.625 0.607
Decision tress (C4.5) 0.975 0.703 0.964
Support vector machine (SVM) 0.977 0.703 0.992
AdaBoosting 0.972 0.719 0.898
Bagging 0.975 0.703 0.954
MLP for cost-sensitive classification (NNCS) 0.496 0.819 0.113
SMOTE + MLP 0.722 0.724 0.177
NCR + CPSO 0.894 0.827 0.694
NCR + MLP 0.642 0.751 0.144
Flat ensemble of ANN 0.958 0.732 0.725
Ensemble of ANN using (NCL) 0.9710.803 0.814