CareNet: Linking Home-router Network Traffic to DSM-5 Depressive Behavior Indicators

TL;DR

CareNet tackles privacy-preserving mental-health sensing by deriving DSM-5 MDD indicators from router header metadata with edge processing. The core FASL fusion combines network metrics into daily criterion likelihoods $L_k(t)$ through signed components $S_{k,b}(t)$ and a DSM-style gate with window parameters $M$, $N$, and threshold $\theta$. The system aggregates indicator signals into episode-level decisions consistent with the DSM-5 rule (at least 5 criteria with a core criterion). The evaluation on realistic traces demonstrates ability to detect altered sleep timing and attentional patterns without payload inspection, highlighting a path toward auditable, privacy-preserving digital phenotyping in homes.

Abstract

Digital mental-health sensing increasingly depends on mobile or wearable devices that require intrusive permissions and continuous user compliance. We present CareNet, a router-centric system that transforms household network metadata into interpretable behavioral indicators aligned with DSM-5 depressive-symptom domains. All processing occurs locally at the home gateway, preserving privacy while maintaining visibility of temporal routines. The core contribution is the Fuzzy Additive Symptom Likelihood (FASL), a transparent formulation that fuses header-level metrics into daily criterion-level likelihoods using bounded fuzzy memberships and additive aggregation. Combined with a DSM-style temporal gate, FASL integrates short-term traffic fluctuations into persistent, clinically interpretable indicators. Evaluation on realistic multi-day traces shows that CareNet captures characteristic patterns such as delayed sleep timing and attentional instability without payload inspection. The results highlight the feasibility of reproducible, explainable behavioral inference from router-side telemetry.

Figures (17)

• Figure 1: Overview of CareNet's approach: From parsing network traffic and extracting features to mapping features into BOMs and DSM-aligned indicators. CareNet's source-code is available: nef_app_pcap_to_dsm5_public
• Figure 2: Sankey diagram illustrating CareNet's general workflow
• Figure 3: CareNet detailed workflow, detailing key components and the flow through parsing, storage, feature extraction, and BOM aggregation toward DSM-5 MDD indicators and an interpretable gate.
• Figure 4: Each square represents one day in the last $M$ days. A day is green if the daily likelihood for criterion $k$ exceeds the threshold ($L_k(d)\ge\theta$), otherwise red. Count the green (positive) days and if at least $N$ of the last $M$ days are positive, the criterion is marked present.
• Figure 7: Daily criterion likelihoods for C8 and C4 after FASL aggregation and DSM-style gating.