Wrapper-Naive Bayes Approach to Perform Efficient Customer Behavior Prediction

Performing efficacious customer behavior analysis makes the enterprises to understand the potential and unpotential consumers with firms and further assist to enhance enterprise's business and customer retention. To comprehend the in-depth pattern of customer behavior in this work, the NB approach is carried out. However, real-world customer data collected comprises of uncertainties variables (redundant) which makes the NB algorithm degrades its performance in customer prediction due to its conditional independence presumption. To address the above problem, attribute selection procedure is carried out to choose the flawless variable subgroup to model with the NB algorithm. In this work, two distinct variable selection approaches are carried out (sequential forward selection-NB and genetic algorithm-NB) and compare the experimental outcome obtained with the standard Naive Bayes, averaged one-dependence estimators, correlation-based feature weighting NB, NBTREE, and DTNB. Experimental results divulge the variable selection with NB achieves higher accuracy of 86.8116 and 86.087 customer prediction compared to other existing approaches like standard Naive Bayes, averaged one-dependence estimators, correlation-based feature weighting NB, NBTREE, and DTNB. From the experimental outcome, we can witness that the proposed methodology is superior compared to standard NB and other existing approaches.