Prédicteur Fuzzy Rule Guided Machine Learning Approach for Retail Sales Demand Prediction

Abstract

The retail business has become more competitive more than ever with the rising global rivalry and technological advancements. Hence overseeing the future demand of products plays a crucial role in retail market to achieve corporate objectives while gratifying clients. As demand of products are inclined by various socio-economic factors, it is a fundamental task to identify their sway to maximize the turnover while curtailing perils. Often if the demand for products is not captured timely it will adversely distress the overall enactment of retail business. Though this long held problem is widely addressed in current ambiance from different perspectives, an approach which considers wide solution space out of several possible simulations is still in study. Proposed solution takes diverse set of factors into account based on past sales, product properties, economic rates and expert’s knowledge as input features. Feature selection was performed using a filter of correlation coefficients against target and minimized input space was used to train extensive set of supervised machine learning algorithms. The hyper parameterization of individual learners was established using a grid of broadest possible intuitive range. The ideal machine learning algorithm with best possible parameters was then selected by wrapper based fuzzy rule learner which takes several performance measures of individual models into account after cross validation. The subjective solution proposes an end to end seamless machine learning pipeline to predict retail sales demand starting from natural set of input features with a competitive predictive performance of its kind.