Hybrid Feature Selection Approach for Naive Bayes to Improve Consumer Behavior Analysis

Abstract

Consumer behavior research in the banking firms is critical in recognizing consumer expectations as well as identifying potential risk customers of bank enterprises. Since access to consumer data is challenging at the beginning of the day and time-consuming. But in recent days due to advancement in internet technology the collection of data related to customer and products are enormous. With the efficient analysis of customer data lead to better opportunities in decision making and also to recognize high-risk customer profile that can reduce the risk of loss by taking action into account. In this research, a machine learning methodology is applied to conduct consumer analysis using Naïve Bayes. But due to the presence of redundant, missing and noisy variables in data sets, Naïve Bayes can perform poorly in the prediction of performance. In order to eliminate the correlated and unnecessary attributes in the dataset and to enhance model efficiency, a hybrid feature selection (HFS-IGFS) approach is applied to get the best optimal feature subset for modeling with Naïve Bayes. The experiment procedure is conducted using bank datasets obtained from UCI repository and results are compared between the naïve Bayes with (Filter, wrapper and HFS approach) and Naïve Bayes without HFS. The experimental results reveal HFS chooses best subset of attributes with reduced computational time and also increase in NB performance prediction is achieved.