Title: An unsupervised learning approach to basket type definition in FMCG sector based on household panel data

Authors: Ahmet Talha Yigit; Tolga Kaya; Utku Dogruak

Addresses: Department of Industrial Engineering, Istanbul Technical University, 34367, Istanbul, Turkey ' Department of Management Engineering, Istanbul Technical University, 34367, Istanbul, Turkey ' Ipsos Turkey, 34854, Istanbul, Turkey

Abstract: The purpose of this study is to propose a clustering-based modelling approach to define the main groups of baskets in Turkish fast-moving consumer goods (FMCG) industry regarding the sectoral decomposition, the total value and the size of the baskets. To do this, based on the information regarding nearly three million basket purchases made in 2018 by more than 14,000 households, alternative unsupervised learning methods such as K-means, and Gaussian mixtures are implemented to obtain and define the basket patterns in Turkey. Additionally, a supervised ensemble learning approach based on XGBoost method is also selected among fully connected neural networks and random forest models to assign the new baskets into the existing clusters. Results show that, 'SaveTheDay', 'CareTrip', 'Breakfast', 'SuperMain' and 'MeatWalk' are among the most important basket types in Turkish FMCG sector.

Keywords: basket analysis; cluster analysis; K-means; fast-moving consumer goods; FMCG; supervised learning; consumer panel; ensemble learning; deep learning.

DOI: 10.1504/IJIDS.2022.125187

International Journal of Information and Decision Sciences, 2022 Vol.14 No.3, pp.243 - 259

Received: 15 Sep 2020
Accepted: 15 Oct 2021
Published online: 01 Sep 2022 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article