Abstract
Understanding human lifestyle patterns is essential for personalized services and urban planning, yet traditional tracking methods like GPS pose significant privacy and battery consumption challenges. This paper introduces a framework to infer individual lifestyles—specifically mobility patterns and domestic habits—using only anonymized WiFi scan lists. By converting these scans into mobility graphs and applying temporal semantic labeling, the study demonstrates that meaningful behavioral insights can be extracted without geographic coordinates. Analysis of a large-scale dataset shows that temporal patterns of stay duration are highly indicative of place function, allowing for the identification of "Home" and "Work" locations with high probability.
Introduction
The digital footprint left by smartphones offers a window into human behavior. Historically, researchers have relied on Global Positioning System (GPS) data or Call Detail Records (CDRs) to map these behaviors. However, GPS is computationally expensive and raises severe privacy concerns due to its precision. CDRs, while more private, often lack the temporal and spatial granularity required for detailed lifestyle analysis.
This research addresses the "privacy-preserving lifestyle discovery" problem. The authors propose using anonymized WiFi Access Point (AP) IDs—data that is frequently collected but often overlooked—to reconstruct daily routines. The challenge lies in the lack of geographic context: an anonymized ID tells you a user is near a specific device, but not whether that device is in a residential or commercial zone.
Methodology
The proposed framework utilizes a multi-stage pipeline to transform raw, anonymized logs into structured lifestyle metrics.
Stay Place Extraction and Mobility Graphs
The first step involves identifying "stay places"—locations where a user remains for at least 10 minutes. This is achieved using the UIM clustering algorithm, which groups similar WiFi fingerprints to represent a single physical location. Once these places are established, a weighted undirected mobility graph is constructed for each user. In this graph, nodes represent stay places, and edges represent the frequency of transitions between them.
Activity Area Detection and Semantic Labeling
To identify broader "activity areas" (e.g., a work district), the researchers apply community detection algorithms, such as edge betweenness and modularity, to the mobility graphs.
A key innovation of this work is the semantic labeling of these areas without GPS. Each significant location is represented as a 24-dimensional vector, where each dimension corresponds to the average stay duration for a specific hour of the day. By applying K-means clustering to these vectors, the system can distinguish between "Home" (high occupancy from 9 PM to 6 AM) and "Work" (high occupancy from 9 AM to 6 PM).
Dataset Analysis
The study leverages the Device Analyzer dataset from the University of Cambridge, encompassing over 17,000 participants across 150 countries. To ensure the reliability of the lifestyle discovery, the researchers filtered for users with at least six months of continuous data, resulting in a core group of 746 users.
As shown in the figure above, the majority of WiFi scanlists are relatively short, typically containing fewer than 10 unique APs. This sparsity necessitates robust clustering algorithms to accurately define stay places.
The log duration analysis highlights the necessity of the filtering process. While the dataset is massive, many users contribute data for only a short period. By focusing on the long-term "tail" of the distribution—those with over six months of logs—the authors ensure that the inferred lifestyles reflect stable, long-term habits rather than transient behaviors.
Key Results and Insights
The analysis revealed several distinct patterns in human behavior:
- Home and Work Regularity: 72% of users spend between 8 and 14 hours at home daily. Workplace stays are notably more "regular" than home stays, showing lower standard deviation in duration, which reflects the structured nature of professional schedules.
- The "Homebody" Phenomenon: Approximately 5.8% of users were found to stay home for over 20 hours a day, likely representing remote workers or individuals with limited mobility.
- Mobility Metrics: The researchers introduced "Mobility Activeness" (average daily transitions) and "Mobility Diversity" (variety of non-essential areas visited). Most users (70%) exhibit an activeness score between 2 and 4, indicating a routine centered around a few primary locations.
Conclusion and Future Work
This study demonstrates that anonymized WiFi data is a powerful, privacy-respecting alternative to GPS for lifestyle discovery. The primary limitation remains the lack of ground truth for validation, as the data is entirely anonymized. Future research aims to integrate non-sensitive sensors, such as accelerometers, to further refine the semantic labeling of stay places and validate these heuristic models with smaller, labeled control groups.
Summary updated