Enhancing the Performance of the Classifiers for Customer Churn Analysis in Telecommunication Data using EMOTE

Customer Churn is the term refers to the customers who are in threat to leave the company. Growing number of such customers are becoming critical for the telecommunication sector and the telecom sector are also in a situation to retain them to avoid the revenue loss. Prediction of such behaviour is very essential for the telecom sector and Classifiers proved to be the effective one for the same. A well balanced dataset is a vital resource for the classifiers to yield the best prediction. All existing classifiers tend to perform poor on imbalanced dataset. An imbalanced dataset is the one, where the classification attribute is not evenly distributed. The cause of poor performance on such dataset is that, the classifiers look for overall accuracy not by taking into account of the relative distribution of each class. Like the other real time applications, the telecommunication churn application also has the class imbalance problem. So it is extremely vital to go for fine balanced dataset for classification. In this paper, an empirical method EC_for TELECAM (Enhanced Classifier for TELEcommunication Churn Analysis Model) using EMOTE (Enhanced Minority Oversampling TEchnique) has been proposed to improve the performance of the classifier for customer churn analysis in telecom dataset. The key idea of the proposed model is that, fine-tuning the misclassified instances into correctly classified instances using their nearest neighbour. To evaluate the performance of the proposed method, Different UCI repository datasets are used with different ratios of imbalance. The experimental result shows that, the proposed method effectively improves the performance of the classifier, through which it extracts the best decision rule for the prediction. In order to perform the Churn analysis, the primary data with 235 samples was collected through structured questionnaire. To extract the decision rule for the churn predicting system, the proposed method was executed on the collected data. In order to prove that, the extracted rules are statistically significant, Discriminant Factor Analysis using SPSS is also carried out. The evaluation results show that, the extracted rules to predict the customer churn are most significant with the related attributes. As an overall, the experimental results show that, the proposed method outperformed and extremely improve the accuracy of the classifier by which it able to achieve the best prediction over the Customer Churn.