Evaluating the potential of machine learning for promotional demand planning and price optimisation in retail

Date
2022-04
Journal Title
Journal ISSN
Volume Title
Publisher
Stellenbosch : Stellenbosch University
Abstract
ENGLISH ABSTRACT: Retailers face numerous challenges due to the highly competitive environment of the retail sector. In order to remain financially sustainable and competitive, retailers often make use of processes such as demand forecasting and promotion- or price planning. Though the literature includes various methods of demand forecasting and price optimisation, many studies have failed to regard the impact of product category information in a machine learning context. The current study extends the existing literature by including the performance of machine learning for demand forecasting and price optimisation in complex retail environments. The current study regards the performance of machine learning when forecasting SKU-level promotional demand as well as category-wide profit maximisation by leveraging important intra-category information. The study comprises the sales and pricing data of three distinct product categories, each of which includes four focal products. The demand forecasting approach consisted of developing four boosting algorithms where product promotion and intra- category information is considered. The demand models were developed using a LASSO feature selection procedure in combination with a Bayesian hyperparameter optimisation process. For the price optimisation model a Deep-Q Network (DQN) was developed using a 4-layer neural network. The pricing optimisation model was developed to make product category pricing decisions for a 12-week horizon, while using the optimal demand model as input. Based on the empirical results, the optimal demand model was the LGBM algorithm which outperformed the other models in every forecasting scenario. For price optimisation, the Deep-Q Network model was able to act as intended and improve the product category profits by simulating a. complex decision-making process.
AFRIKAANSE OPSOMMING: Klemhandelaars staar talle mtdagmgs m die geslg as gevolg van cue hoogs mededmgende omgewing van die kleinhandelsektor. 0m finansieel volhoubaar en mededingend te bly, maak kleinhandelaars dikwels gebruik van prosesse soos vraagvoorspelling en promosie- of prysbeplanning. Alhoewel die literatuur verskeie metodes van vraagvoorspelling en prysoptimalisering insluit, het baie studies versuim om die impak van produkkategoriey inligting in 'n masjienleerkonteks te beskou. Die huidige studie brei die bestaande literatuur uit deur die prestasie van masjienleer vir vraagvoorspelling en prysoptimalisering in komplekse kleinhandelomgewings in te sluit. Die huidige studie ag die prestasie van masjienleer by die voorspelling van SKU-vlak promosievraag sowel as kategoriewye winsmaksimering deur belangrike intra-kategorie inligting te benut. Die studie bestaan uit die verkoops en prysdata van drie afsonderlike produkkategorieé, wat elk Vier fokusprodukte insluit. Die vraagvoorspellingsbenadering het bestaan uit die ontwikkeling van Vier 'boosting' algoritmes waar produkpromosie en intrakategorie- inligting oorweeg word. Die vraagmodelle is ontwikkel deur gebruik te maak van 'n LASSO kenmerkseleksieprosedure in kombinasie met 'n Bayesian hiperparameter optimeringsproses. Vir die prysoptimeringsmodel is 'n Deep-Q Network (DQN) ontwikkel met behulp van 'n 4-laag neurale netwerk. Die prysoptimeringsmodel is ontwikkel om produkkategorieyprysbesluite vir 'n 12-week-horison te neem, waar die optimale aanvraagmodel as inset gebruik word. Gebaseer op die empiriese resultate, was die optimale aanvraagmodel die LGBNI-algoritme wat beter as die ander modelle in elke voorspellingscenario gevaar het. Vir prysoptimering kon die Deep-Q Network-model optree soos verwag en die produkkategorie-winste verbeter deur 'n komplekse besluitnemingsproses te simuleer.
Description
Thesis (MEng)--Stellenbosch University, 2022.
Keywords
Product promotion, UCTD, Demand (Economic theory) -- Forecasting, Machine learning, Price policy, Industrial, Reinforcement learning
Citation