Machine Learning Approach for Mobility Context Classification Using Radio Beacons

The study of human mobility becomes more and more crucial these days in transportation studies, urban planning, crowd mobility behaviors, and even more. In this paper, we propose a novel approach for studying human mobility by building a light machine learning (ML) model using observation of wireless networking information from WiFi and Bluetooth low energy (BLE) that are today naturally present in everyday devices such as mobile phones. Our goal is to build a mobility classification system using communicating devices of any kind with low processing complexity. However, we propose a new approach for mobility classification using a real dataset of WiFi and BLE beacons collected over one year for around 90 hours in different scenarios and conditions. The first model (B-model) aims to identify the status of a device if stationary or mobile. Then a complementary model (M-model) is applied to determine a more precise real-life situation of the device, which could be a Home, Office, Bus, Train, etc. The results show that decision-tree-based ensemble ML algorithms like LGBMClassifier and XGBClassifier gave the best results, in terms of accuracy and f1 score for both models with an accuracy of 99% and 94% respectively, confirming the capability of classifying mobility context from only WiFi and BLE data. We believe that such an approach could be leveraged for studying human mobility and an important step towards the large deployment of mobility-based applications by leveraging everyday mobile phones.