Using machine learning to explore the impact of weather on high street retail in the UK

Authors: Natalie Rose*, University of Liverpool, Les Dolega, University of Liverpool
Topics: Spatial Analysis & Modeling, Quantitative Methods, Climatology and Meteorology
Keywords: High street retail, consumer behaviour, weather, tree-based modelling
Session Type: Paper
Day: 4/4/2019
Start / End Time: 1:10 PM / 2:50 PM
Room: Congressional A, Omni, West
Presentation File: No File Uploaded

Retail is one of the most important economic sectors in the UK, on that is highly dynamic, and the nature of which has undergone significant changes in recent years, predominantly due to the rise of e-commerce and convenience culture. As a result, an understanding of the key influences on product sales is arguably amongst the most valuable information for retailers to help them stay competitive in the current retail landscape. Weather is a key influential factor on both product demand and consumer purchasing decisions, both directly and indirectly, however due to the complexity of consumer behaviours the impact that weather has on sales can be particularly difficult to monitor.

This study uses daily sales data from a retailer in the UK, combined with meteorological data, to study the varying impact of different weather types on product sales and identify the categories that are most susceptible to influence from changing weather conditions. In addition, spatial variations in weather impact will be explored at both local and regional levels. A random forest methodology is employed to undertake this analysis and quantify the impact that weather conditions have on sales. Furthermore, partial dependence plots are used to enable the relationships between weather variables and sales to be examined. By combining these machine learning methods with expansive sales datasets, it is hoped that we can provide a greater understanding of the relationship between retail and weather, which will subsequently aid retailers in the reduction of waste and economic losses.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login