Prediction of Retail Prices Using Local Competitors
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
One of many applications of data mining is the prediction of retail prices. Often the price of a product at a store depends on the price of the same product at other stores. Our focus is to predict two types of prices, i.e., the start of the day price of the product at each store in a city and the real time (current) price of the product at each store in a city. We focus on predicting the product prices by identifying competitors of a store, for various notions of competitors. We also consider relevant information that could help us improve the prediction accuracy e.g., wholesale prices, and present strategies on how to use this relevant information in the model. We present di erent strategies of variable selection and training within the Vector Autoregressive Model (VAR) in order to accurately predict the product price. We evaluate our strategies of identifying competitors, use of the relevant information, variable selection and training of the VAR model by calculating the Average Price Di erence (APD) and comparing it with the APD obtained from simple Autoregression which uses no competitors or any relevant information. We use the data set provided to us by a crowdsourcing application for a time period of almost 3 years for ve North American cities. The results show that using the competitors and relevant knowledge does improve the accuracy by 40% - 50% for the start of the day price prediction and around 13% - 40% for the real time price prediction.