Predictive performance measurement system for retail industry using neuro-fuzzy system based on swarm intelligence

• Main Authors: Holimchayachotikul P., Leksakul K.
• Format: Journal
• Published: 2017
• Online Access: https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85016079917&origin=inward; http://cmuir.cmu.ac.th/jspui/handle/6653943832/40584
• Institution: Chiang Mai University

© 2016, Springer-Verlag Berlin Heidelberg. Between 2011 and 2013, convenience store retail business grew dramatically in Thailand. As a result, most companies have increasingly been choosing the application of performance measurement systems. This significantly results in poor performance measurement regarding future business lagging measure. To solve this problem, this research presents a hybrid predictive performance measurement system (PPMS) using the neuro-fuzzy approach based on particle swarm optimization (ANFIS-PSO). It is constructed from many leading aspects of convenience store performance measures and projects the competitive level of future business lagging measure. To do so, monthly store performance measures were first congregated from the case study value chains. Second, data cleaning and preparations by headquarter accounting verification were carried out before the proposed model construction. Third, these results were used as the learning dataset to derive a predictive performance measurement system based on ANFIS-PSO. The fuzzy value of each leading input was optimized by parallel processing PSO, before feeding to the neuro-fuzzy system. Finally, the model provides a future performance for the next month’s sales and expense to managers who focused on managing a store using desirability function (D i ). It boosted the sales growth in 2012 by ten percentages using single PPMS. Additionally, the composite PPMS was also boosted by the same growth rate for the store in the blind test (July 2013–February 2014). From the experimental results, it can be concluded that ANFIS-PSO delivers high-accuracy modeling, delivering much smaller error and computational time compared to artificial neural network model and supports vector regression but its component searching time differs significantly because of the complexity of each model.