AN ENHANCED BANK CUSTOMERS CHURN PREDICTION MODEL USING A HYBRID GENETIC ALGORITHM AND K-MEANS FILTER AND ARTIFICIAL NEURAL NETWORK

Abstract

Customer churn prediction is an important issue in banking industry and has gained attention over the years. Early identification of customers likely to leave a bank is vital in order to retain such customers. Predicting churning is a data mining tasks that require several data mining approaches. Churn prediction based on Artificial Neural Networks (ANNs) have been successful, however, they are affected by the noise or outliers present in such datasets. The effect of such noise, and number of training samples on churn prediction was investigated. Two filters were applied to the data, the Genetic Algorithm (GA) and Kmeans filter. The filtered data were used to train an ANN model and tested with a 30% unfiltered data. The performance show that the training performance improved when noise was filtered while the testing performance was affected by the unbalanced data caused by filtering