Abstract—Over the years, many different importance-performance analysis (IPA) variations have emerged as it is a primary tool for analyzing customer satisfaction. One of the recent IPA variations is back-propagation neural network based importance-performance analysis (BPNN based IPA) that utilizes BPNN to measure Importance. To investigate the performance of the BPNN based IPA, the authors compared two types of BPNN models that have one and multiple output neurons referred as BPNN (regression) and BPNN (classification) respectively, with multiple linear regression (MLR). This comparison demonstrates that the BPNN (regression) does not outperform MLR in term of model accuracy and training time, yet BPNN (classification) is superior to MLR and BPNN (regression) in term of model accuracy and predictive power. This finding leads to a reconsideration of the BPNN model used in the present BPNN based IPA.

Index Terms—Back-propagation neural network, empirical comparison, importance-performance analysis.

Dept. of Electronics and Computer Science, University of Southampton, Southampton, United Kingdom.

Cite:Boonyarat Phadermrod, Richard M. Crowder, Gary B. Wills, "Attribute Importance Measure Based on Back-Propagation Neural Network: An Empirical Study," International Journal of Computer and Electrical Engineering vol. 7, no. 2, pp. 118-127, 2015.