Understanding Customers and Their Grouping via WiFi Sensing for Business Revenue Forecasting

Abstract

Emerging technologies provide a variety of sensors in smartphones for state monitoring. Among all the sensors, the ubiquitous WiFi sensing is one of the most important components for the use of Internet access and other applications. In this work, we propose a WiFi-based sensing for store revenue forecasting by analyzing the customers’ behavior, especially the grouped customers’ behavior. Understanding customers’ behavior through WiFi-based sensing should be beneficial for selling increment and revenue improvement. In particular, we are interested in analyzing the customers’ behavior for customers who may visit stores together with their partners or they visit stores with similarly patterns, called group behavior or group information for store revenue forecasting. The proposed method is realized through a WiFi signal collecting AP which is deployed in a coffee shop continuously for a period of time. Following a procedure of data collection, preprocessing, and feature engineering, we adopt Support Vector Regression to predict the coffee shop’s revenue, as well as other useful information such as the number of WiFi-using devices, the number of sold products. Overall, we achieve as good as $7.63\%$, $11.32\%$ and $14.43\%$ in the prediction on the number of WiFi-using devices, the number of sold products and the total revenue respectively if measured in Mean Absolute Percentage Error (MAPE) from the proposed method in its peak performance. Moreover, we have observed an improvement in MAPE when either the group information or weather information is included.