Interpretable machine learning for demand modeling with high-dimensional data using Gradient Boosting Machines and Shapl
- PDF / 621,814 Bytes
- 10 Pages / 595.276 x 790.866 pts Page_size
- 48 Downloads / 206 Views
RESEARCH ARTICLE
Interpretable machine learning for demand modeling with high‑dimensional data using Gradient Boosting Machines and Shapley values Evgeny A. Antipov1 · Elena B. Pokryshevskaya1 Received: 10 January 2020 / Accepted: 16 February 2020 © Springer Nature Limited 2020
Abstract Forecasting demand and understanding sales drivers are one of the most important tasks in retail analytics. However, traditionally, linear models and/or models with a small number of predictors have been predominantly used in sales modeling. Taking into account that real-world demand is naturally determined by complex substitution and complementation patterns among a large number of interrelated SKUs, nonlinear effects of prices, promotions, seasonality, as well as many other factors, their lagged values, and interactions, a realistic model has to be able to account for all that. We propose a conceptual model for sales modeling based on standard POS data available to any retailer and generate almost 500 potentially useful predictors of a focal SKU’s sales accordingly. In our comparison of three classes of models, Gradient Boosting Machines outperformed Random Forests and Elastic nets. By using interpretable machine learning methods, we came up with actionable insights related to the importance of various groups of predictors from the conceptual model, as well as demonstrated how helpful it can be for marketing managers to decompose predictions into the effects of individual regressors by using an approximation of Shapley values for feature attribution. Keywords Sales forecasting · Shapley value · Interpretable machine learning · Random forest · Gradient Boosting Machines · Elastic net
Introduction Quantifying price sensitivity of sales as well as true effectiveness of promotions remain challenging problems due to the complexity of relationships that determine sales. For example, nonpromoted products may be cannibalized by promoted ones, the effectiveness of a promotion is likely to depend on promotional activity in previous weeks, joint promotions of certain products may have a synergy effect on some other product’s sales, etc. As a result, either none or only a portion of the sales lift generated by a promotion is incremental for retailers (Gedenk 2018). More specifically, it has been shown that more than 50% of all promotions were not profitable for the retailer (Ailawadi et al. 2007). Even if a promotion is beneficial for a manufacturer, it is likely to * Evgeny A. Antipov [email protected] 1
National Research University Higher School of Economics, Kantemirovskaya St. 3, Saint‑Petersburg, Russia 194100
cause a negative impact for a retailer because switching happens from nonpromoted items to promoted ones that typically have lower margin. To manage promotions effectively, one needs to evaluate and quantify the effects that promotions have on sales, which is why a good sales response model is needed. One of the classical models used in literature on mathematical models in marketing for sales forecasting as well as price and
Data Loading...
