Machine Learning assisted Gross Weight Prediction for Alcoholic Beverages in Retail

Abstract

In this paper we present a machine learning pipeline developed specifically for the product group of alcoholic beverages with focus on the two segments wine and beer which constitute the major part of a retailer’s alcoholic beverages inventory. We focus on exploiting expert knowledge about the data domain to engineer features tailored to prediction of the important attribute gross weight. Experiments with data from a major retail company show that our proposed machine learning approach with feature enriched data achieves superior results which are more robust than those obtained by traditional heuristic approaches on the original data. In practical terms this is a step towards fully automated product data generation and maintenance reducing manual effort and thus costs for a retail company.