Business performance forecasting of convenience store based on enhanced fuzzy neural network

Reliable business performance forecasting of convenience store (CVS) can not only help in making the correct local selection decision but also in decreasing the store cost and thereby enlarging the profit significantly. Therefore, the main aim of this paper is to design an enhanced fuzzy neural network (EFNN)-based predictor to forecast the business performance of CVS. Without considering relevant domain knowledge, traditional fuzzy neural networks suffer from the problem of low accuracy of forecasting unseen examples. Moreover, traditional fuzzy neural networks have to turn weights with a kind of time-consuming gradient steepest descent training algorithm. Considering the relationship between the evaluation factors globally, we devise the EFNN which assigns connection weights based on the expert domain knowledge without painstakingly and repeatedly turning them. Furthermore, by generating and refining the activation function based on genetic algorithm, our EFNN can provide comprehensive and accurate activation functions and fit a wider range of nonlinear models. By experimenting our methods with several benchmark methods, the proposed methods are found to have an optimal accuracy in forecasting the business performance of CVS with a permanent good performance.