arxiv: 2605.01616 · v1 · submitted 2026-05-02 · 💻 cs.LG · cs.AI· cs.CY· cs.NI

Recognition: unknown

From Packets to Patterns: Interpreting Encrypted Network Traffic as Longitudinal Behavioral Signals

Rameen Mahmood , Omar El Shahawy , Souptik Barua , Zachary Beattie , Jeffrey Kaye , Xuhai "Orson'' Xu , Chao-Yi Wu , Danny Yuxing Huang

Authors on Pith no claims yet

Pith reviewed 2026-05-09 14:34 UTC · model grok-4.3

classification 💻 cs.LG cs.AIcs.CYcs.NI

keywords encrypted network trafficbehavioral sensinglongitudinal analysistransformer modelssparse autoencoderssleep disturbancestressloneliness

0 comments

The pith

Encrypted smartphone network traffic captures within-person changes in sleep, stress, and loneliness using learned representations.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper tests whether encrypted smartphone network traffic can passively capture behavioral patterns related to sleep disturbance, stress, and loneliness. It models shared structure with a transformer backbone and per-user adapters, then applies a sparse autoencoder to extract interpretable features for distinct activity patterns. These features are related to self-reported outcomes through generalized estimating equations with Mundlak decomposition, which separates stable between-person differences from within-person changes over time. The analysis shows stress linked mainly to between-person differences, loneliness to within-person variation, and sleep disturbance to both, with the learned features uniquely revealing the within-person dynamics missed by predefined traffic metrics.

Core claim

Encrypted smartphone network traffic contains longitudinal behavioral signals for sleep, stress, and loneliness. A transformer backbone with per-user adapters models both typical individual behavior and deviations from it, while a sparse autoencoder extracts distinct behavioral features from the traffic. Relating these to self-reported outcomes using generalized estimating equations with Mundlak decomposition reveals that stress associates with stable between-person differences, loneliness with within-person variation, and sleep disturbance with a combination of both, and that these within-person dynamics are not captured by predefined network-traffic features.

What carries the argument

Transformer backbone with per-user adapters to represent individual baselines and deviations, followed by a sparse autoencoder to extract interpretable behavioral features from encrypted traffic patterns.

If this is right

Stress is primarily associated with stable between-person differences in network traffic patterns.
Loneliness is associated with within-person variation in traffic patterns over time.
Sleep disturbance is associated with a combination of between-person differences and within-person changes.
Learned representations detect within-person behavioral dynamics that predefined network-traffic features do not capture.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The approach could enable passive detection of shifts away from a person's typical network activity patterns as indicators of changing loneliness levels.
Applications might involve using deviations in daily traffic to flag potential increases in sleep disturbance or stress for further monitoring.
The separation of between- and within-person signals suggests the method could track both stable traits and temporary states in behavior.

Load-bearing premise

Associations between the learned traffic representations and self-reported behavioral outcomes reflect genuine behavioral signals rather than confounding factors such as device type, network conditions, or usage habits, and that findings generalize beyond the studied sample.

What would settle it

A replication in a new sample where the extracted features show no remaining association with within-person changes in self-reported loneliness after controlling for device type and network conditions would falsify the claim that learned representations capture genuine behavioral signals.

Figures

Figures reproduced from arXiv: 2605.01616 by Chao-Yi Wu, Danny Yuxing Huang, Jeffrey Kaye, Omar El Shahawy, Rameen Mahmood, Souptik Barua, Xuhai "Orson'' Xu, Zachary Beattie.

**Figure 1.** Figure 1: Overview of the proposed framework. Stage 1: data collection and feature construction from encrypted network view at source ↗

**Figure 2.** Figure 2: Two-phase training procedure for the shared transformer backbone and per-user residual adapters. Purple boxes view at source ↗

**Figure 3.** Figure 3: SAE feature selection pipeline. From 512 learned features, 364 are active; 70 pass the cross-user generality filter; 18 view at source ↗

**Figure 5.** Figure 5: (Top) Robust SAE feature–outcome associations. view at source ↗

**Figure 6.** Figure 6: Feature c1 behavioral profile, capturing midday messaging and social media activity. (a) Hour-of-day activation view at source ↗

read the original abstract

Human behavior is difficult to observe continuously at scale, yet it leaves measurable traces in everyday device use. We test whether encrypted smartphone network traffic -- a ubiquitous, always-on, passive sensing modality -- can passively capture behavioral patterns related to sleep, stress, and loneliness. We model shared behavioral structure using a transformer backbone with per-user adapters, allowing the model to represent both typical individual behavior and deviations from it. To make these representations interpretable, we apply a sparse autoencoder to extract behavioral features corresponding to distinct patterns of activity. We relate these features to sleep disturbance, stress, and loneliness using generalized estimating equations with Mundlak decomposition, separating between-person differences from within-person changes over time. We find that the three outcomes reflect distinct temporal structures: stress is primarily associated with stable between-person differences, loneliness with within-person variation, and sleep disturbance with a combination of both. Notably, these within-person dynamics are not captured by predefined network-traffic features, demonstrating the value of learned representations for longitudinal behavioral sensing. These results establish encrypted network traffic as a viable passive sensing modality, revealing interpretable behavioral dynamics -- particularly deviations from an individual's baseline -- that are not visible in raw traffic features.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper shows learned features from encrypted traffic can track within-person shifts in sleep, stress, and loneliness that hand-crafted stats miss, but the abstract supplies almost no numbers or checks to judge whether the associations are real.

read the letter

The core contribution is a pipeline that feeds encrypted smartphone traffic into a transformer with per-user adapters, runs a sparse autoencoder on the embeddings, and then applies Mundlak decomposition inside GEE models to separate stable between-person differences from within-person changes over time. They report that stress links mostly to between-person structure, loneliness to within-person variation, and sleep to both, and that these within-person signals do not appear when they swap in predefined traffic features instead. That contrast is the clearest new claim. The setup is sensible for a passive, always-on modality that already exists on devices, and the adapter plus sparse autoencoder route is a reasonable way to get both personalization and some interpretability without starting from scratch. The longitudinal framing and the explicit within/between split are also useful moves that prior traffic-sensing work has not emphasized as much. Credit for trying to make the representations legible rather than leaving them as opaque vectors. The main weakness is that the abstract gives no sample size, no number of users or observation days, no validation details, no error bars, and no description of how they handled obvious confounds such as device model, OS version, carrier, or background app behavior. Without those, it is impossible to tell whether the within-person coefficients reflect genuine behavioral deviations or leftover correlations with stable usage habits. The stress-test concern about residual device or network effects is on point; the abstract does not mention sensitivity analyses or exhaustive covariate sets for them. If the full paper has solid controls and the results survive them, the work is worth following. If not, the central contrast collapses. This is aimed at researchers in digital phenotyping, passive sensing, and privacy-aware health monitoring. Anyone already working on network-traffic features for behavioral signals will see a concrete next step. It deserves peer review because the idea is fresh and the within-person angle matters, even though the current write-up needs substantial methodological expansion and robustness checks before it can be trusted.

Referee Report

2 major / 1 minor

Summary. The manuscript claims that encrypted smartphone network traffic can be used as a passive, always-on sensing modality to capture longitudinal behavioral patterns related to sleep disturbance, stress, and loneliness. It models shared structure with a transformer backbone augmented by per-user adapters, extracts interpretable features via sparse autoencoders, and relates the resulting representations to self-reported outcomes through generalized estimating equations that incorporate Mundlak decomposition to separate between-person and within-person effects. The central findings are that the three outcomes exhibit distinct temporal structures (stress primarily between-person, loneliness within-person, sleep a mix) and that these within-person dynamics are invisible to predefined network-traffic features, thereby demonstrating the value of learned representations.

Significance. If the empirical associations hold after proper controls, the work is significant because it establishes encrypted traffic as a viable, privacy-preserving source for behavioral sensing at scale. The combination of per-user adapters and sparse autoencoders provides a concrete route to interpretable longitudinal signals, and the Mundlak-GEE analysis supplies a statistically grounded distinction between stable traits and time-varying deviations. This could influence future passive-sensing pipelines in digital health and psychology by showing that learned representations can surface signals missed by hand-crafted features.

major comments (2)

[Methods] Methods (modeling and statistical analysis sections): The pipeline description does not specify the full covariate set used in the Mundlak GEE models (e.g., device type, OS version, carrier, background app patterns, or network conditions) nor report sensitivity checks that would confirm the per-user adapters and SAE features remove stable confounders. Without these, the claim that within-person coefficients reflect genuine behavioral deviations (rather than residual device/usage effects) cannot be evaluated and is load-bearing for both the temporal-structure results and the contrast with predefined features.
[Results] Results (comparison with predefined features): The assertion that within-person dynamics are not captured by predefined network-traffic features is presented without a quantitative head-to-head evaluation (e.g., a table of GEE coefficient magnitudes, model-fit statistics, or cross-validated predictive performance between SAE-derived features and the hand-crafted baseline set). This omission makes the key novelty claim difficult to assess.

minor comments (1)

[Abstract] Abstract: Key quantitative details such as participant count, observation period, or effect-size ranges are absent; adding one sentence summarizing these would improve the reader's ability to gauge the scale of the reported associations.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which identify key areas where additional methodological detail and quantitative comparisons will strengthen the manuscript. We address each point below and have revised the paper accordingly.

read point-by-point responses

Referee: [Methods] Methods (modeling and statistical analysis sections): The pipeline description does not specify the full covariate set used in the Mundlak GEE models (e.g., device type, OS version, carrier, background app patterns, or network conditions) nor report sensitivity checks that would confirm the per-user adapters and SAE features remove stable confounders. Without these, the claim that within-person coefficients reflect genuine behavioral deviations (rather than residual device/usage effects) cannot be evaluated and is load-bearing for both the temporal-structure results and the contrast with predefined features.

Authors: We agree that explicit documentation of the covariate set and sensitivity analyses is required to support the interpretation of within-person effects. In the revised manuscript we will expand the statistical analysis subsection to enumerate all covariates entered into the Mundlak GEE models (device type, OS version, carrier, background app activity, and network-condition indicators). We will also add a dedicated sensitivity section that reports (i) models with and without the per-user adapters, (ii) models that replace SAE features with raw traffic statistics, and (iii) checks for residual device-level confounding. These additions will allow readers to assess whether the reported within-person coefficients are robust to stable device and usage factors. revision: yes
Referee: [Results] Results (comparison with predefined features): The assertion that within-person dynamics are not captured by predefined network-traffic features is presented without a quantitative head-to-head evaluation (e.g., a table of GEE coefficient magnitudes, model-fit statistics, or cross-validated predictive performance between SAE-derived features and the hand-crafted baseline set). This omission makes the key novelty claim difficult to assess.

Authors: We accept that a direct quantitative comparison is necessary to substantiate the claim. The revised Results section will include a new table that reports, for each outcome, the magnitude and significance of within-person coefficients obtained from (a) the SAE-derived feature set and (b) the predefined network-traffic feature set. The table will also present model-fit statistics (quasi-likelihood information criterion) and, where appropriate, cross-validated predictive performance. This side-by-side evaluation will make the differential sensitivity of the two feature representations transparent. revision: yes

Circularity Check

0 steps flagged

No significant circularity; standard pipeline applied to external data

full rationale

The paper trains a transformer backbone with per-user adapters on encrypted traffic, extracts interpretable features via sparse autoencoder, and associates them with self-reported outcomes using GEE plus Mundlak decomposition. These steps use external behavioral labels and perform empirical contrasts against predefined features; no equation or claim reduces by construction to its own inputs, fitted parameters renamed as predictions, or load-bearing self-citations. The within-person vs. between-person separation follows directly from the Mundlak formulation applied to longitudinal data rather than definitional equivalence.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Based solely on the abstract, no explicit free parameters, axioms, or invented entities are identifiable; the work relies on standard machine learning components and established statistical techniques whose assumptions are not detailed here.

pith-pipeline@v0.9.0 · 5548 in / 1251 out tokens · 36805 ms · 2026-05-09T14:34:47.804867+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

79 extracted references · 4 canonical work pages · 1 internal anchor

[1]

Fadel Adib, Hongzi Mao, Zachary Kabelac, Dina Katabi, and Robert C Miller. 2015. Smart homes that monitor breathing and heart rate. InProceedings of the 33rd annual ACM conference on human factors in computing systems. 837–846

2015
[2]

Apple Inc. [n. d.]. Categories and Discoverability. Apple Developer Documentation. https://developer.apple.com/app-store/categories/ Accessed: April 2026

2026
[3]

Apple Services Performance Partners. 2026. iTunes Search API. https://performance-partners.apple.com/search-api Accessed: 2026-04-22

2026
[4]

Noah Apthorpe, Danny Yuxing Huang, Dillon Reisman, Arvind Narayanan, and Nick Feamster. 2018. Keeping the smart home private with smart (er) iot traffic shaping.arXiv preprint arXiv:1812.00955(2018)

work page arXiv 2018
[5]

Jakob E Bardram. 2022. Software architecture patterns for extending sensing capabilities and data formatting in mobile sensing.Sensors 22, 7 (2022), 2813

2022
[6]

Scott Barnett, Kit Huckvale, Helen Christensen, Svetha Venkatesh, Kon Mouzakis, and Rajesh Vasa. 2019. Intelligent sensing to inform and learn (InSTIL): a scalable and governance-aware platform for universal, smartphone-based digital phenotyping for research and clinical applications.Journal of medical Internet research21, 11 (2019), e16399

2019
[7]

Yoav Benjamini and Yosef Hochberg. 1995. Controlling the false discovery rate: a practical and powerful approach to multiple testing. Journal of the Royal statistical society: series B (Methodological)57, 1 (1995), 289–300

1995
[8]

Tjeerd W Boonstra, Jennifer Nicholas, Quincy JJ Wong, Frances Shaw, Samuel Townsend, and Helen Christensen. 2018. Using mobile phone sensor technology for mental health research: integrated analysis to identify hidden challenges and potential solutions.Journal of medical Internet research20, 7 (2018), e10131

2018
[9]

Trenton Bricken, Adly Templeton, Joshua Batson, Brian Chen, Adam Jermyn, Tom Conerly, Nick Turner, Cem Anil, Carson Denison, Amanda Askell, et al. 2023. Towards monosemanticity: Decomposing language models with dictionary learning.Transformer Circuits Thread2, 5 (2023), 6

2023
[10]

Andrew T Campbell, Shane B Eisenman, Nicholas D Lane, Emiliano Miluzzo, Ronald A Peterson, Hong Lu, Xiao Zheng, Mirco Musolesi, Kristóf Fodor, and Gahng-Seop Ahn. 2008. The rise of people-centric sensing.IEEE Internet Computing12, 4 (2008), 12–21

2008
[11]

Luca Canzian and Mirco Musolesi. 2015. Trajectories of depression: unobtrusive monitoring of depressive states by means of smartphone mobility traces analysis. InProceedings of the 2015 ACM international joint conference on pervasive and ubiquitous computing. 1293–1304

2015
[12]

David Cella, William Riley, Arthur Stone, Nan Rothrock, Bryce Reeve, Susan Yount, Dagmar Amtmann, Rita Bode, Daniel Buysse, Seung Choi, et al. 2010. The Patient-Reported Outcomes Measurement Information System (PROMIS) developed and tested its first wave of adult self-reported health outcome item banks: 2005–2008.Journal of clinical epidemiology63, 11 (20...

2010
[13]

Sheldon Cohen, Tom Kamarck, and Robin Mermelstein. 1983. A global measure of perceived stress.Journal of health and social behavior (1983), 385–396

1983
[14]

Hoagy Cunningham, Aidan Ewart, Logan Riggs, Robert Huben, and Lee Sharkey. 2023. Sparse autoencoders find highly interpretable features in language models.arXiv preprint arXiv:2309.08600(2023)

work page internal anchor Pith review arXiv 2023
[15]

Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. InProceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, volume 1 (long and short papers). 4171–4186

2019
[16]

Jason A Donenfeld. 2017. Wireguard: next generation kernel network tunnel.. InNDSS. 1–12

2017
[17]

Afsaneh Doryab, Daniella K Villalba, Prerna Chikersal, Janine M Dutcher, Michael Tumminia, Xinwen Liu, Sheldon Cohen, Kasey Creswell, Jennifer Mankoff, John D Creswell, et al. 2019. Identifying behavioral phenotypes of loneliness and social isolation with passive sensing: statistical analysis, data mining and machine learning of smartphone and fitbit data...

2019
[18]

Nelson Elhage, T Hume, Catherine Olsson, N Schiefer, T Henighan, S Kravec, Z Hatfield-Dodds, R Lasenby, D Drain, C Chen, et al. 2022. Toy models of superposition. Transformer Circuits Thread, 2022

2022
[19]

2006.OpenVPN: Building and integrating virtual private networks

Markus Feilner. 2006.OpenVPN: Building and integrating virtual private networks. Packt Publishing Ltd

2006
[20]

BSB Gonçalves, Taísa Adamowicz, Fernando Mazzilli Louzada, Claudia Roberta Moreno, and John Fontenele Araujo. 2015. A fresh look at the use of nonparametric analysis in actimetry.Sleep medicine reviews20 (2015), 84–91. •Mahmood et al

2015
[21]

Gireesh K Gupta. 2011. Ubiquitous mobile phones are becoming indispensable.ACM Inroads2, 2 (2011), 32–33

2011
[22]

Unsoo Ha, Sohrab Madani, and Fadel Adib. 2021. WiStress: Contactless stress monitoring using wireless signals.Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies5, 3 (2021), 1–37

2021
[23]

Ellen L Hamaker, Rebecca M Kuiper, and Raoul PPP Grasman. 2015. A critique of the cross-lagged panel model.Psychological methods 20, 1 (2015), 102

2015
[24]

Gabriella M Harari, Nicholas D Lane, Rui Wang, Benjamin S Crosier, Andrew T Campbell, and Samuel D Gosling. 2016. Using smartphones to collect behavioral data in psychological science: Opportunities, practical considerations, and challenges.Perspectives on Psychological Science11, 6 (2016), 838–854

2016
[25]

Gabriella M Harari, Sandrine R Müller, Min SH Aung, and Peter J Rentfrow. 2017. Smartphone sensing methods for studying behavior in everyday life.Current opinion in behavioral sciences18 (2017), 83–90

2017
[26]

2025.Phone Screen Time Statistics

Harmony Healthcare IT. 2025.Phone Screen Time Statistics. https://www.harmonyhit.com/phone-screen-time-statistics/ Accessed: 2026-04-20

2025
[27]

Louise C Hawkley and John T Cacioppo. 2010. Loneliness matters: A theoretical and empirical review of consequences and mechanisms. Annals of behavioral medicine40, 2 (2010), 218–227

2010
[28]

Afzal Hossain and Christian Poellabauer. 2016. Challenges in building continuous smartphone sensing applications. In2016 IEEE 12th international conference on wireless and mobile computing, networking and communications (WiMob). IEEE, 1–8

2016
[29]

Chen-Yu Hsu, Yuchen Liu, Zachary Kabelac, Rumen Hristov, Dina Katabi, and Christine Liu. 2017. Extracting gait velocity and stride length from surrounding radio signals. InProceedings of the 2017 CHI conference on human factors in computing systems. 2116–2126

2017
[30]

Tianrui Hu, Daniel J Dubois, and David Choffnes. 2023. Behaviot: Measuring smart home iot behavior using network-inferred behavior models. InProceedings of the 2023 ACM on internet measurement conference. 421–436

2023
[31]

Thomas R Insel. 2017. Digital phenotyping: technology for a new science of behavior.Jama318, 13 (2017), 1215–1216

2017
[32]

Nicholas C Jacobson, Berta Summers, and Sabine Wilhelm. 2020. Digital biomarkers of social anxiety severity: digital phenotyping using passive smartphone sensors.Journal of medical Internet research22, 5 (2020), e16875

2020
[33]

Stephen Kent and Karen Seo. 2005. RFC 4301: Security architecture for the Internet protocol

2005
[34]

Liu Kesheng, Ni Yikun, Li Zihan, and Duan Bin. 2020. Data mining and feature analysis of college students’ campus network behavior. In2020 5th IEEE International Conference on Big Data Analytics (ICBDA). IEEE, 231–237

2020
[35]

Philipp Krieter and Andreas Breiter. 2018. Analyzing mobile application usage: generating log files from mobile screen recordings. In Proceedings of the 20th international conference on human-computer interaction with mobile devices and services. 1–10

2018
[36]

Nicholas D Lane, Emiliano Miluzzo, Hong Lu, Daniel Peebles, Tanzeem Choudhury, and Andrew T Campbell. 2010. A survey of mobile phone sensing.IEEE Communications magazine48, 9 (2010), 140–150

2010
[37]

Neal Lathia, Kiran K Rachuri, Cecilia Mascolo, and Peter J Rentfrow. 2013. Contextual dissonance: Design bias in sensor-based experience sampling methods. InProceedings of the 2013 ACM international joint conference on Pervasive and ubiquitous computing. 183–192

2013
[38]

Hansoo Lee, Joonyoung Park, and Uichin Lee. 2022. A systematic survey on android api usage for data-driven analytics with smartphones. Comput. Surveys55, 5 (2022), 1–38

2022
[39]

Yingcheng Liu, Guo Zhang, Christopher G Tarolli, Rumen Hristov, Stella Jensen-Roberts, Emma M Waddell, Taylor L Myers, Meghan E Pawlik, Julia M Soto, Renee M Wilson, et al . 2022. Monitoring gait at home with radio waves in Parkinson’s disease: A marker of severity, progression, and medication response.Science Translational Medicine14, 663 (2022), eadc9669

2022
[40]

Yongsen Ma, Gang Zhou, and Shuangquan Wang. 2019. WiFi sensing with channel state information: A survey.ACM Computing Surveys (CSUR)52, 3 (2019), 1–36

2019
[41]

Rameen Mahmood, Tousif Ahmed, Sai Teja Peddinti, and Danny Yuxing Huang. 2025. Large Language Models for Real-World IoT Device Identification.arXiv preprint arXiv:2510.13817(2025)

work page arXiv 2025
[42]

R Mahmood, A David, D Hu, N Alshurafa, LM Haux, J Hester, A Kiselica, S Liu, C Qiu, CY Wu, et al. 2026. Digital Phenotyping via Passive Network Traffic Monitoring: Prospective Observational Study in University Students.JMIR Formative Research(2026)

2026
[43]

Tarek Mahmud, Meiru Che, and Guowei Yang. 2022. Android api field evolution and its induced compatibility issues. InProceedings of the 16th ACM/IEEE International Symposium on Empirical Software Engineering and Measurement. 34–44

2022
[44]

Tyler McDonnell, Baishakhi Ray, and Miryung Kim. 2013. An empirical study of api stability and adoption in the android ecosystem. In 2013 IEEE International Conference on Software Maintenance. IEEE, 70–79

2013
[45]

Markus Miettinen, Samuel Marchal, Ibbad Hafeez, Nadarajah Asokan, Ahmad-Reza Sadeghi, and Sasu Tarkoma. 2017. Iot sentinel: Automated device-type identification for security enforcement in iot. In2017 IEEE 37th international conference on distributed computing systems (ICDCS). IEEE, 2177–2184

2017
[46]

Geoffrey Miller. 2012. The smartphone psychology manifesto.Perspectives on psychological science7, 3 (2012), 221–237

2012
[47]

David C Mohr, Mi Zhang, and Stephen M Schueller. 2017. Personal sensing: understanding mental health using ubiquitous sensors and machine learning.Annual review of clinical psychology13 (2017), 23–47

2017
[48]

Yair Mundlak. 1978. On the pooling of time series and cross section data.Econometrica: journal of the Econometric Society(1978), 69–85. From Packets to Patterns: Interpreting Encrypted Network Traffic as Longitudinal Behavioral Signals•

1978
[49]

Stephan Nef and Bruno Rodrigues. 2025. CareNet: Linking Home-router Network Traffic to DSM-5 Depressive Behavior Indicators. arXiv preprint arXiv:2511.12772(2025)

work page arXiv 2025
[50]

Subigya Nepal, Wenjun Liu, Arvind Pillai, Weichen Wang, Vlado Vojdanovski, Jeremy F Huckins, Courtney Rogers, Meghan L Meyer, and Andrew T Campbell. 2024. Capturing the college experience: a four-year mobile sensing study of mental health, resilience and behavior of college students during the pandemic.Proceedings of the ACM on interactive, mobile, wearab...

2024
[51]

Jukka-Pekka Onnela, Caleb Dixon, Keary Griffin, Tucker Jaenicke, Leila Minowada, Sean Esterkin, Alvin Siu, Josh Zagorsky, and Eli Jones. 2021. Beiwe: A data collection platform for high-throughput digital phenotyping.Journal of Open Source Software6, 68 (2021), 3417

2021
[52]

Jukka-Pekka Onnela and Scott L Rauch. 2016. Harnessing smartphone-based digital phenotyping to enhance behavioral and mental health.Neuropsychopharmacology41, 7 (2016), 1691–1696

2016
[53]

2025.Mobile Fact Sheet

Pew Research Center. 2025.Mobile Fact Sheet. https://www.pewresearch.org/internet/fact-sheet/mobile/ Accessed: 2026-04-20

2025
[54]

Malik Muhammad Qirtas, Evi Zafeiridi, Dirk Pesch, and Eleanor Bantry White. 2022. Loneliness and social isolation detection using passive sensing techniques: scoping review.JMIR mHealth and uHealth10, 4 (2022), e34638

2022
[55]

Byron Reeves, Nilam Ram, Thomas N Robinson, James J Cummings, C Lee Giles, Jennifer Pan, Agnese Chiatti, MJ Cho, Katie Roehrick, Xiao Yang, et al. 2021. Screenomics: A framework to capture and analyze personal life experiences and the ways that technology shapes them.Human–Computer Interaction36, 2 (2021), 150–201

2021
[56]

Daniel W Russell. 1996. UCLA Loneliness Scale (Version 3): Reliability, validity, and factor structure.Journal of personality assessment 66, 1 (1996), 20–40

1996
[57]

Sohrab Saeb, Mi Zhang, Christopher J Karr, Stephen M Schueller, Marya E Corden, Konrad P Kording, and David C Mohr. 2015. Mobile phone sensor correlates of depressive symptom severity in daily-life behavior: an exploratory study.Journal of medical Internet research 17, 7 (2015), e4273

2015
[58]

Ramona Schoedel and Matthias R Mehl. 2024. Mobile sensing methods.HT Reis, T. West, & CM Judd (Eds.), Handbook of research methods in social and personality psychology(2024), 297–321

2024
[59]

Ramona Schoedel, Thomas Reiter, Michael D Krämer, Yannick Roos, Markus Bühner, David Richter, Matthias R Mehl, and Cornelia Wrzus. 2026. Person-related selection bias in mobile sensing research: Robust findings from two panel studies.Journal of Personality and Social Psychology(2026)

2026
[60]

Rahul Anand Sharma, Elahe Soltanaghaei, Anthony Rowe, and Vyas Sekar. 2022. Lumos: Identifying and localizing diverse hidden {IoT}devices in an unfamiliar environment. In31st USENIX Security Symposium (USENIX Security 22). 1095–1112

2022
[61]

Christopher Slade, Yinan Sun, Wei Cheng Chao, Chih-Chun Chen, Roberto M Benzo, and Peter Washington. 2025. Current challenges and opportunities in active and passive data collection for mobile health sensing: a scoping review.JAMIA open8, 4 (2025), ooaf025

2025
[62]

Vincent F Taylor, Riccardo Spolaor, Mauro Conti, and Ivan Martinovic. 2016. Appscanner: Automatic fingerprinting of smartphone apps from encrypted network traffic. In2016 IEEE European Symposium on Security and Privacy (EuroS&P). IEEE, 439–454

2016
[63]

2024.Scaling monosemanticity: Extracting interpretable features from claude 3 sonnet

Adly Templeton. 2024.Scaling monosemanticity: Extracting interpretable features from claude 3 sonnet. Anthropic

2024
[64]

Alisha Ukani, Ariana Mirian, and Alex C Snoeren. 2021. Locked-in during lock-down: undergraduate life on the internet in a pandemic. InProceedings of the 21st ACM Internet Measurement Conference. 480–486

2021
[65]

Aditya Vaidyam, John Halamka, and John Torous. 2022. Enabling research and clinical use of patient-generated health data (the mindLAMP Platform): digital phenotyping study.JMIR mHealth and uHealth10, 1 (2022), e30557

2022
[66]

Eus JW Van Someren, Dick F Swaab, Christopher C Colenda, Wayne Cohen, W Vaughn McCall, and Peter B Rosenquist. 1999. Bright light therapy: improved sensitivity to its effects on rest-activity rhythms in Alzheimer patients by application of nonparametric methods. Chronobiology international16, 4 (1999), 505–518

1999
[67]

Rui Wang, Fanglin Chen, Zhenyu Chen, Tianxing Li, Gabriella Harari, Stefanie Tignor, Xia Zhou, Dror Ben-Zeev, and Andrew T Campbell
[68]

In Proceedings of the 2014 ACM international joint conference on pervasive and ubiquitous computing

StudentLife: assessing mental health, academic performance and behavioral trends of college students using smartphones. In Proceedings of the 2014 ACM international joint conference on pervasive and ubiquitous computing. 3–14

2014
[69]

Sinan Wang, Yibo Wang, Xian Zhan, Ying Wang, Yepang Liu, Xiapu Luo, and Shing-Chi Cheung. 2022. Aper: evolution-aware runtime permission misuse detection for android apps. InProceedings of the 44th International Conference on Software Engineering. 125–137

2022
[70]

Shweta Ware, Chaoqun Yue, Reynaldo Morillo, Jin Lu, Chao Shang, Jayesh Kamath, Athanasios Bamis, Jinbo Bi, Alexander Russell, and Bing Wang. 2018. Large-scale automatic depression screening using meta-data from wifi infrastructure.Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies2, 4 (2018), 1–27

2018
[71]

Ari Winbush, Daniel McDuff, John Hernandez, Andrew Barakat, Allen Jiang, Conor Heneghan, Benjamin W Nelson, and Nicholas B Allen. 2025. Smartphone use in a large US adult population: Temporal associations between objective measures of usage and mental well-being.Proceedings of the National Academy of Sciences122, 43 (2025), e2427311122

2025
[72]

2026.tshark(1) – Dump and analyze network traffic

Wireshark Foundation. 2026.tshark(1) – Dump and analyze network traffic. Wireshark Project. https://www.wireshark.org/docs/man- pages/tshark.html TShark manual page, Wireshark documentation. •Mahmood et al

2026
[73]

Wil Witting, IH Kwa, Pieter Eikelenboom, Majid Mirmiran, and Dick F Swaab. 1990. Alterations in the circadian rest-activity rhythm in aging and Alzheimer’s disease.Biological psychiatry27, 6 (1990), 563–572

1990
[74]

Wei Xuan, Meghna Roy Chowdhury, Yi Ding, and Yixue Zhao. 2025. Unlocking Mental Health: Exploring College Students’ Well- being through Smartphone Behaviors. In2025 IEEE/ACM 12th International Conference on Mobile Software Engineering and Systems (MOBILESoft). IEEE, 66–70

2025
[75]

Yuzhe Yang, Yuan Yuan, Guo Zhang, Hao Wang, Ying-Cong Chen, Yingcheng Liu, Christopher G Tarolli, Daniel Crepeau, Jan Bukartyk, Mithri R Junna, et al. 2022. Artificial intelligence-enabled detection and assessment of Parkinson’s disease using nocturnal breathing signals.Nature medicine28, 10 (2022), 2207–2215

2022
[76]

Siamak Yousefi, Hirokazu Narui, Sankalp Dayal, Stefano Ermon, and Shahrokh Valaee. 2017. A survey on behavior recognition using WiFi channel state information.IEEE Communications Magazine55, 10 (2017), 98–104

2017
[77]

Chaoqun Yue, Shweta Ware, Reynaldo Morillo, Jin Lu, Chao Shang, Jinbo Bi, Jayesh Kamath, Alexander Russell, Athanasios Bamis, and Bing Wang. 2020. Automatic depression prediction using internet traffic characteristics on smartphones.Smart Health18 (2020), 100137

2020
[78]

George Zerveas, Srideepika Jayaraman, Dhaval Patel, Anuradha Bhamidipaty, and Carsten Eickhoff. 2021. A transformer-based framework for multivariate time series representation learning. InProceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining. 2114–2124

2021
[79]

# changed

Mingmin Zhao, Shichao Yue, Dina Katabi, Tommi S Jaakkola, and Matt T Bianchi. 2017. Learning sleep stages from radio signals: A conditional adversarial architecture. InInternational conference on machine learning. PMLR, 4100–4109. From Packets to Patterns: Interpreting Encrypted Network Traffic as Longitudinal Behavioral Signals• A Participant Inclusion A...

2017