Improving Time Use Measurement with Personal Big Data Collection – The Experience of the European Big Data Hackathon 2019

Mattia Zeni, Ivano Bison, Fernando Reis, Britta Gauckler, Fausto Giunchiglia

TL;DR

This paper introduces i-Log, a smartphone-based system designed to modernize Time Use Surveys (TUS) by combining active user reporting with passive sensor data collection. By leveraging personal "big data" from smartphones, the authors demonstrate a path toward "Smart Statistics" that reduces participant burden while significantly increasing data granularity and accuracy.

What Problem Does This Paper Address?

Traditional Time Use Surveys are the bedrock of sociological and economic planning, but they are currently facing a crisis of relevance. Historically, these surveys rely on paper diaries where participants manually record activities every 10 minutes. This method is:

  1. High Burden: It is exhausting for participants, leading to high dropout rates.
  2. Infrequent: Due to high costs and complexity, they are often only conducted once every decade.
  3. Inaccurate: They rely on human memory and struggle to capture the nuances of modern life, such as multitasking or time spent on digital media.

The authors aim to bridge this gap by using ubiquitous smartphone technology to automate data collection and provide a more "synchronic" view of human behavior.

Methods

The core of the research is the i-Log system, a dual-mode data collection platform:

  • Active Mode: The app sends hourly notifications asking users four key questions: What are you doing? Where are you? Who are you with? and What is your mood? To reduce cognitive load, answers are mapped to a standardized ontology based on WordNet.
  • Passive Mode: In the background, i-Log collects data from over 20 hardware and software sensors, including accelerometers, GPS, Wi-Fi connection states, battery levels, and even "Doze Mode" status.

Technically, the system utilizes a cloud-based backend with a Cassandra database for linear scalability and Apache Spark for distributed data processing. A key methodological innovation is the "one-click-deployment" pattern, allowing researchers to spin up the entire infrastructure instantly for new pilot studies.

Datasets

The paper describes datasets provided during the European Big Data Hackathon 2019:

  • i-Log Dataset: The primary novel dataset, featuring 66 active participants who contributed ~190,000 hours of sensor data and 8,548 validated diary entries over 14 days.
  • HETUS (Harmonised European Time Use Survey): Traditional paper-based micro-data from five countries used as a baseline.
  • MOTUS (Modular Online Time Use Survey): An online TUS dataset from a sample of teachers in Flanders.
  • myBigO: Data from an external research project on childhood obesity, providing sensor signals and food-related self-reports.

Key Insights and Ideas

  • The Duality of Data: The "aha moment" is that active diary entries serve as "ground truth" labels for passive sensor data. This allows machine learning models to learn a user's habits and eventually "fill the gaps" when a user forgets to answer a prompt.
  • The "Burden" Sweet Spot: Initial testing found that prompting users twice per hour led to high dropout rates. The researchers settled on an hourly intervention as the optimal balance between data granularity and user retention.
  • Hardware as Proxy for Behavior: One hackathon team (the Greek team) discovered they could predict sleep patterns with high accuracy simply by monitoring the phone’s "Doze Mode" and rest states, rather than requiring a manual "I am sleeping" entry.
  • Privacy through Separation: To handle the sensitive nature of geolocation data, the system separates identifying information (like emails) from the sensor streams, providing a multi-level privacy preservation framework.

Key Results

The system was stress-tested during the 2019 European Big Data Hackathon, where 17 national teams competed:

  • Adoption: 6 out of 17 teams successfully integrated i-Log data into their prototypes.
  • Predictive Success: Teams built models to predict transportation modes (foot vs. bus) and identify "zones of mobility" (home vs. work) using geo-positioning data.
  • Scalability: The system successfully handled an average of 500MB of uncompressed data per smartphone per day, proving it can scale to larger national studies.
  • Award-Winning Prototypes: The winning team (Statistics Poland) created an open-source dashboard for population time-use analysis based on these "smart" data streams.

Limitations and Future Work

The authors acknowledge several hurdles:

  • OS Constraints: At the time of the study, i-Log was Android-only, which introduces sampling bias. An iOS version is currently under development.
  • Battery Life: While optimized, continuous background sensing remains a challenge for older devices.
  • Data Sparsity: Some machine learning approaches (like CNNs for physical activity) struggled due to the relatively small sample size of the pilot study.
  • The "Black Box" of Automation: There is a lingering challenge in validating the "completeness" of data when automatic systems malfunction or when users provide inconsistent labels.

Future work will focus on integrating data from wearable devices (smartwatches) via Bluetooth to capture physical activity even when the user isn't carrying their phone.

Summary updated

Abstract

This article assesses the experience with i-Log at the European Big Data Hackathon 2019, a satellite event of the New Techniques and Technologies for Statistics (NTTS) conference, organised by Eurostat. i-Log is a system that enables capturing personal big data from smartphones’ internal sensors to be used for time use measurement. It allows the collection of heterogeneous types of data, enabling new possibilities for sociological urban field studies. Sensor data such as those related to the location or the movements of the user can be used to investigate and gain insights into the time diaries’ answers and assess their overall quality. The key idea is that the users’ answers are used to train machine-learning algorithms, allowing the system to learn from the user’s habits and to generate new time diaries’ answers. In turn, these new labels can be used to assess the quality of existing ones, or to fill the gaps when the user does not provide an answer. The aim of this paper is to introduce the pilot study, the i-Log system and the methodological evidence that emerged during the survey.
BibTeX
@inproceedings{zeni2021improving,
  author = {Mattia Zeni and Ivano Bison and Fernando Reis and Britta Gauckler and Fausto Giunchiglia},
  title = {{Improving Time Use Measurement with Personal Big Data Collection – The Experience of the European Big Data Hackathon 2019}},
  booktitle = {RePEc:},
  year = {2021},
  doi = {10.2478/jos-2021-0015},
}

Loading related content...

Figure