Full Program »
Features weight estimation using genetic algorithm for customer churn prediction in telecom sector
Customers are the key source of revenue and their retention is of utmost importance for businesses to grow. Customer Churn (losing customers) is regarded as a growing concern generally in most of the service based companies and specifically in the telecommunication sector. In order to retain the customer to increase the profit, many, data mining based, Customer Churn Prediction (CCP) techniques have been developed. These techniques are mostly focused on the attribute reduction and domain-based weighting in classification for improving the classifier’s accuracy. Thus overlooking the fact that by attribute reduction there is a high chance of loss of important data. Similarly, attribute domain based weighting requires the domain expert to assign the weight to all attributes making it a subjective exercise. Therefore, there is a need of a technique to self-assign more appropriate weights without involving domain expert, losing the data or feature reduction. This paper presents a novel features weighting technique. The technique employs genetic algorithm to incorporate a benchmark for CCP based on Naïve Bayes (NB) classification which relaxes the attribute independence assumption and frequency estimator without losing any data/attribute from the original set. Experiments have been conducted on publically available datasets and compared the performance of the proposed approach applied on the existing simple NB. The experimental results have demonstrated that the proposed technique out-performed when compared to simple NB and achieved overall 89.1% accuracy and 95.65% precision which shows the effectiveness and efficiency of the proposed learning benchmark.