Negative Correlation Learning for Customer Churn Prediction: A Comparison Study

Table 5

Tuning parameters for data mining techniques used in the comparison study.

Method	Parameters
GP	Population size = 1000,; Maximum number of generations = 100, functions = {*, /, −, +, IF, , ,; tree max depth = 10, tree max length = 30; elites = 1, selection mechanism = tournament selection crossover point probability = 90%, mutation = probability 15%
ANN with BP	Activation function = Sigmoid, Epoches = 5000, Learning Rate = 0.3, Momentum = 0.2
SVM	Cost = 1, Gamma = 10000
IBK	Number of neighbours = 1, nearest neighbor search algorithm = linear search (brute force search)
AdaBoost	Number of classifiers = 10
Bagging	Number of classifiers = 10
NNCS	Hidden layers = 2, hidden nodes = 15
SMOTE	Number of neighbors = 5
NCR + CPSO	Number of neighbors = 5 for SMOTE, number of particles = 75 for CPSO