Improved Churn Prediction Based on Supervised and Unsupervised Hybrid Data Mining System

Retaining a customer plays a vital role in success of every sales firm. Not only it increases the profit margin of the firm but also it maintains the ranking of the firm in the competitive market. Every organization competes for the success of itself in the competitive market, so it aims to retain the existing customers through customer preservation techniques. Because preserving an existing customer is less costly than adding new customer. Hence, customer association management (CAM) and customer maintenance (CM) are two important parameters in determining the success of the organization. Also quality service is another parameter which reduces the customer churn to a greater extent. Hence, every organization conducts a customer churn forecast as a valuable step because it aims at customer maintenance and organization success. In this work instead of single classifier resulting in low efficiency, hybrid supervised and unsupervised techniques are deployed to achieve improved churn prediction. In stage one, data cleaning is carried out to eliminate deviations in data set. In stage two, clustering algorithms such as K-Means and K-Medoids are used to group customers with similar trends. In stage three, hold-out methods based training and testing data sets are obtained from the above clusters. In the next stage, the training and testing are carried out by algorithms such as decision tree (DT), k-nearest neighbor (KNN), support vector machine (SVM), linear discriminant analysis (LDA), and naive bayes (NB). In the final stage, sensitivity, specificity, and accuracy are measured to evaluate the efficiency of the hybrid system.