Auric-Grid/E1.S6-Adolescent-Behavioral-Health-Dataset

Dataset README: Adolescent Social Withdrawal & Depression — Synthetic Research Dataset

Version: 1.0 | Patients: 1,000 | Longitudinal visits: 12,912
Institution: Auric Grid Laboratory

---

Overview

This is a synthetic, research-grade dataset modelling social withdrawal and depressive symptomatology in adolescents (ages 12–19). It is structured for multimodal machine learning tasks including cross-sectional classification, longitudinal trajectory prediction, relapse risk modelling, and clinical intervention analysis.

The dataset is split into two conceptual parts and four files:

File	Part	Type	Rows	Columns
partA_crosssectional.csv	Part A	Cross-sectional (full features)	1,000	37
static_features-5.csv	Part A (subset)	Static features only	1,000	23
partB_longitudinal.csv	Part B	Longitudinal with archetype labels	12,912	29
dynamic_features-5.csv	Part B (subset)	Dynamic/visit-level features only	12,912	28

All 1,000 patients appear in both Part A and Part B. The longitudinal files contain 8–18 visits per patient (median 13).

---

Part A — Cross-Sectional Data

partA_crosssectional.csv

Complete baseline snapshot per patient. Contains all demographic, psychosocial, behavioural, and clinical variables collected at enrolment.

Key fields:

Column	Type	Description
Patient_ID	str	Unique patient identifier (hex string)
Age	int	Age at enrolment (12–19 years)
Sex	str	Female (58.8%), Male (41.2%)
Ethnicity	str	Categorical ethnicity label
School_Level	str	High-school, Middle, Pre-college, Vocational
Socioeconomic_Status	str	Low / Medium / High
Academic_Performance	int	Score 1–10
Attendance_Rate	float	School attendance percentage
Extracurricular_Participation	str	Activity level category
Social_Interaction_Frequency	str	Ordinal frequency label
In_Person_Friendship_Count	int	Count of in-person friends
Bullying_Exposure	str	Type of bullying (NaN = none)
Bullying_Duration_Months	float	Duration in months (NaN if no bullying)
Online_Communication_Hours	float	Daily hours online
Family_History_Mental_Illness	int	Binary flag (0/1)
Adverse_Childhood_Events	float	ACE count score
Family_Conflict_Level	float	Scaled conflict score
Peer_Support_Level	int	Score 1–10
Screen_Time_Hours	float	Daily screen time hours
Gaming_Or_Isolation_Behavior	str	Behavioural pattern category
Smartphone_Usage_Pattern	str	Avoidant / Entertainment / Mixed
Daily_Routine_Consistency	int	Score 1–10
Physical_Activity_Level	str	Sedentary / Moderate / Active
Sleep_Quality	int	Score 1–10
Social_Anxiety_Score	int	Score 1–30
Emotional_Expression_Score	int	Score 1–10
Mood_Variability	int	Score 1–10
Stress_Level	int	Score 1–10
Concentration_Difficulty	int	Score 1–10
Motivation_Level	int	Score 1–10
Anhedonia_Flag	int	Binary flag (0/1)
Counseling_History	str	Type of prior counselling (NaN = none)
PHQ_A_Score	float	PHQ-A total score (0–27)
Depression_Severity_Score	int	Composite severity score
Relapse_Risk_Score	float	Continuous risk score (0–1)
Diagnostic_Uncertainty_Flag	int	Binary flag (0/1)
Social_Withdrawal_Category	str	Primary outcome: Mild / Moderate / Severe / NaN

Outcome distribution (Social_Withdrawal_Category):

Category	N
Mild	353
Moderate	147
Severe	79
Not withdrawn / NaN	421

---

static_features-5.csv

A curated 23-column subset of partA_crosssectional.csv retaining only the stable, time-invariant features. Adds three derived columns not in Part A:

Column	Description
WC_Severity	Ordinal withdrawal severity code (0–3, maps to None/Mild/Moderate/Severe)
PHQ_Band	Ordinal PHQ-A band (0=minimal, 1=mild, 2=moderate, 3=severe)
Bullying_Type	Integer-encoded bullying type (0 = none)

Intended for use as the static feature tensor in multimodal or time-series models (e.g., concatenated with per-visit embeddings).

---

Part B — Longitudinal Data

partB_longitudinal.csv

Visit-level records for all 1,000 patients across 12,912 total visits. Each row is one clinical contact. Includes all dynamic tracking variables plus an Archetype label encoding the patient’s overall clinical trajectory.

Archetype distribution:

Archetype	Visits
CBT_Reintegration	3,451
Academic_Relapse	2,546
Bullying_Responder	2,307
Family_Conflict	2,065
Chronic_Disengaged	1,521
Spontaneous_Recovery	1,022

Key fields:

Column	Type	Description
Patient_ID	str	Links to Part A
Archetype	str	Patient-level trajectory archetype (constant per patient)
Visit_Number	int	Sequential visit index (1–18)
Timestamp_Relative_Days	float	Days since enrolment
Session_Type	str	Outpatient-CBT, School-counselor, Family-therapy, Peer-group, Telehealth, Emergency
Time_Since_Last_Session_Days	float	Gap between visits
Social_Interaction_Trend	str	Ordinal trend label
In_Person_Contact_Count	int	Contacts since last visit
Online_Hours_Trend	float	Change in online hours
PHQ_A_Trend	int	PHQ-A score at this visit
Social_Anxiety_Trend	int	Anxiety score at this visit
Mood_Trend_Score	int	Mood score at this visit
Anhedonia_Trend	int	Anhedonia severity at this visit
Attendance_Rate_Trend	float	School attendance % at this visit
Sleep_Quality_Trend	int	Sleep score at this visit
Lifestyle_Change_Score	float	Composite lifestyle delta
Bullying_Active_Flag	int	Ongoing bullying (0/1)
Risk_Score_Trend	float	Dynamic relapse risk score
Clinical_Status_Change	str	Recovered / Stable / Worsening / etc.
Intervention_Flag	int	New intervention at this visit (0/1)
Counseling_Event	float	Counselling event code
Follow_Up_Recommendation	str	Discharge / Routine / Intensify / Psychiatry-referral / School-reintegration-plan
Temporal_Mask	int	Masking flag for temporal models
Social_Remission_Status	int	Achieved social remission (0/1)
MDD_Episode_Flag	int	Active MDD episode (0/1) — present in ~20.2% of visits
School_Reintegration_Flag	int	School reintegration active (0/1)
Psychiatric_Referral_Flag	int	Psychiatric referral issued (0/1)
Time_To_Social_Remission_Days	float	Days to remission (NaN if not yet achieved)
Relapse_Within_90_Days	int	Target label: relapse within 90 days (0/1) — positive rate ~0.14%

---

dynamic_features-5.csv

Identical to partB_longitudinal.csv minus the Archetype column. Contains the same 12,912 visit rows across 1,000 patients. Intended for models where trajectory archetype should not be provided as input (e.g., archetype prediction tasks, or clean dynamic feature tensors).

---

Suggested Use Cases

Task	Files	Target
Social withdrawal severity classification	partA_crosssectional.csv	Social_Withdrawal_Category
Depression severity regression	partA_crosssectional.csv	PHQ_A_Score / Depression_Severity_Score
Relapse risk prediction (static)	static_features-5.csv	Relapse_Risk_Score
Longitudinal trajectory modelling	partB_longitudinal.csv	Clinical_Status_Change
Archetype classification	dynamic_features-5.csv + static_features-5.csv	Archetype (from Part B)
Relapse prediction (dynamic)	partB_longitudinal.csv	Relapse_Within_90_Days
Time-to-remission survival analysis	partB_longitudinal.csv	Time_To_Social_Remission_Days
Multimodal static + temporal fusion	static_features-5.csv + dynamic_features-5.csv	Any longitudinal target

---

Notes

• All data is fully synthetic. No real patient records are represented.
• Patient_ID is a consistent hex string key across all four files and can be used for merging.
• Missing values (NaN) are structurally meaningful in several columns (e.g., Bullying_Exposure = NaN indicates no bullying; Social_Withdrawal_Category = NaN indicates no diagnosed withdrawal).
• dynamic_features-5.csv and partB_longitudinal.csv share identical rows; the only difference is the presence of Archetype in the latter.
• The Temporal_Mask column in the longitudinal files is provided for compatibility with masked sequence models (e.g., BERT-style pretraining on clinical time series).