Evolutionary neural classification for evaluation of retail stores and decision support

Abstract

The neural network paradigm of learning vector quantization (LVQ) and several enhancements of the standard algorithms have demonstrated improved predictive accuracy when applied to simple 'toy' problems. In this paper, we propose a novel approach of evolutionary optimized LVQ classification applied in real world business decision support. We predict the success of retail outlets of a multinational German company in terms of revenue and profit. The predictions are used to support investment decisions, establishing new stores or closing down existing ones with limited prospective profits. In addition, the predictions provide information to change in-store design or product lines of existing stores. The LVQ networks are trained on data reflecting the macroscopic socio-demographic infrastructure and microscopic in-store aspects of existing outlets. Results of numerous computational experiments in a parallelized PC network are compared with standard neural networks, demonstrating pre-eminent results of the novel method.