ECONOMIC ORDER QUANTITY PREDICTIVE MODEL USING SUPERVISED MACHINE LEARNING FOR INVENTORY MANAGEMENT OF THE FAST-MOVING CONSUMER GOODS DISTRIBUTORS

Abstract

Due to the tough competition that exists today, most distribution companies are in a continuous effort to increase their profits and reduce their costs. An improvement in the distributors’ forecasting process could have significant financial and organizational benefits. This requires decision-makers to have a fast, accurate, and efficient demand forecasting solution to be integrated into their business processes. One of the problems that face the distribution industry is how to control inventory levels by means of accurate demand and economic order quantity (EOQ) prediction. Due to some limitations, EOQ cannot be calculated by the formulas. In this situation, machine learning can help to determine the optimum EOQ. Thus, supervised regression models are chosen as the basic tools for EOQ prediction to reduce the uncertainty and enhance the efficiency, since most traditional statistical methods are incapable of modelling nonlinearities that exist in most real data. This research aimed to optimize fast moving consumer goods (FMCGs) inventory levels through investigating a suitable structure of supervised machine learning algorithms that can be used for predicting EOQ and then evaluating the performance of the selected algorithms. The predictive model was developed through using four machine learning algorithms: linear regression (LR), random forest (RF), boosted decision tree (BDT) and artificial neural network (ANN). These algorithms are evaluated for predicting the distributor’s weekly EOQ in two scenarios; parallel (data is certain and available) and sequential (data is predictable). It was found that BDT and ANN produced relatively accurate results in both scenarios. The research was considered as a single-case study to explore successful demand forecasting strategies that leaders of a small, retail, medical supply business used to increase profitability. Data collection is mainly through semi structured, face-to-face interviews with the top manager and other staff involved in inventory control operations. Sample data set was provided by the case company, which is the leading FMCGs distributors in Egypt, consisting of entries from January 2014 to December 2018. Microsoft Azure cloud-based machine learning platform is used for analyzing data and building the predictive model. The model evaluations and results indicated that the sequential approach was the best methodology, and the weakest one was the model used by the company. A machine learning model with a sequential structure of BDT and NNR algorithms can be considered the most suitable structure, as it shows the best results for prediction. This model resulted in the improvement of the distributor’s KPIs which are “available to promise” and the “operating cash flow” up to 83% and 66% respectively compared to the baseline (company’s results).