Identifying Key Determinants for Consumer Experience in Web-Based Retail Stores Using an Improved Fuzzy-Bayesian Model

Abstract

Fuzzy logic (FL) techniques and Bayesian Networks (BN) modelling are popular machine learning (ML) algorithms that have been used for developing predictive and decision support systems across different domains of science. These two ML techniques are reported to have inherent advantages and better suited for handling incomplete and noisy datasets. Recent research works have also reported the hybridization of these two algorithms in order to take advantage of their unique capabilities, majority of these existing works show a hybridization method based on a simple input to output, whereby the output of the fuzzy logic system is fed as input to the Bayesian system. This method of hybridization is subject to a higher algorithmic complexity as the number of nodes or features increases which leads to increased computational time, also, this hybridization method cannot be applied to existing Bayesian network models. This work therefore presents a method of hybridization which maintains the original structure and parameters of a classical Bayesian network but introduces fuzzy membership values to each node of the Bayesian network. Applied to a case study on consumer experience in online stores, this research identifies the most critical factors influencing consumer experience. Given that consumer experience in e-commerce environments depends on multiple interrelated factors, the proposed hybrid method offers a robust solution to the complexities associated with missing, incomplete, and noisy data. The model is developed using data from open academic questionnaires and can serve as a tool for helping managers of businesses and owners in making business decisions, by performing influence and scenario analysis, business managers can easily tell the extent to which individual factors impact consumer experience.