Project Case Study · Mental Health & Behavioural Analytics

Teen Mental Wellness & Behaviour Analytics

Exploring how behavioural habits, AI usage patterns, lifestyle and social support influence mood and stress levels across a global simulated teen dataset — from raw Excel to a Databricks Medallion pipeline to a published Power BI dashboard and analytical report.

Dataset

Inside Teen Minds — Kaggle

Records Analysed

1,000 student records

Deliverables

Databricks Pipeline · Power BI · Report

Context References

WHO (2025) · RAND (2025) · WHO Europe (2024)

Scroll

01 — The Brief

Problem & Approach

Teen mental health is one of the most urgent conversations of our time. Globally, 1 in 7 adolescents aged 10–19 experiences a mental health condition — accounting for 15% of the global burden of disease in that age group (WHO, 2025). And yet the behavioural patterns that drive emotional variation remain poorly understood, particularly in an era where AI tools have become part of everyday student life.

This project set out to explore a simple but important question: do behavioural habits — lifestyle, sleep, AI usage, social support — actually predict how stressed or positive a teen feels? And does demographic background play a role?

Using the Inside Teen Minds dataset from Kaggle — a richly simulated dataset designed to reflect realistic teen behavioural patterns across 8 countries — I built a complete end-to-end analysis: data preparation in Excel, a full Medallion Architecture pipeline in Azure Databricks, a multi-page Power BI dashboard, and a 10-slide analytical report with verified real-world context from WHO and RAND research.

The findings were counterintuitive — and more interesting for it.

Data note: This analysis uses a simulated dataset (Inside Teen Minds, Kaggle — MIT License). It reflects realistic teen behavioural patterns but does not represent real-world survey data. Findings describe patterns within this dataset only. Real-world context statistics are sourced from WHO (2025) and RAND (2025).

Azure Databricks SQL Power BI Excel Medallion Architecture Delta Lake Power Query Pivot Tables DAX Data Governance

02 — Data Engineering

From Raw CSV to Gold Layer — The Full Pipeline

The pipeline started in Excel — where the real analytical work of creating derived columns happened — before being structured into a full Medallion Architecture in Azure Databricks. Four Bronze tables were uploaded, transformed into a single merged Silver table, then aggregated into nine Gold KPI views that fed directly into Power BI. An early flawed Gold view — one that paired a student's most frequent category label with their whole-month average — was caught and dropped in favour of daily-level correlation testing.

Bronze

4 Raw Tables Uploaded

Excel → Databricks

Raw Kaggle CSV loaded into Excel first
9 new columns derived by analyst: Mood_Grp, StressL_Grp, SupFeeling_Grp, SleHrs_Group, ScrTH_Group, Social_Status, Lifestyle_Score, Lifestyle_Level, Used_AI_Num
AI_Tool_ID joined from AI tools lookup table
4 Bronze Delta tables uploaded to Databricks

Silver

1 Merged Silver Table

Databricks SQL

silver_core_mental_health — single enriched view combining assessment, student dimension, weekday/date, and all derived columns
Rebuilt to merge two earlier separate Silver views into one
Data types validated, nulls handled, categories standardised

Gold

9 Gold KPI Views

Databricks SQL

gold_weekday_trends, gold_ai_tool_usage, gold_weekday_full_pattern — usage and emotional scores by day of week
gold_mood_distribution, gold_stress_distribution — above/below average breakdowns
A flawed early view (pairing partial-month labels with whole-month averages) was found and dropped in favour of daily-level correlation testing

Report

Dashboard & Story

Power BI + Excel

Power BI connected to both Silver (measures, groupings) and Gold (weekday trends) tables
Interactive dashboard with dynamic slicers
KPI tables built via Excel Pivot Tables
Stress benchmark = dataset mean (4.04/10 — analyst-defined 1–10 scale)

Benchmarking methodology: Mood and stress scores were normalised to a 1–10 scale defined by the analyst during data preparation. The dataset mean stress score (4.04) was used as the internal benchmark — students above this were classified as Above Average Stress, those below as Below Average Stress. The mood scale midpoint (5.0) serves as the neutral reference. Observed mean mood of 6.0 is above neutral — indicating moderate-to-positive emotional wellbeing across the dataset.

03 — The Dashboard

Power BI — Interactive Dashboard

Built with dynamic slicers for Gender, Country, Age, Lifestyle Level, Support Feeling, Sleep Group and Screen Time Group — enabling granular exploration of any combination of characteristics. Every finding below is backed by both an aggregate view and, where it matters, a daily-level statistical test.

Overview — KPI cards & weekday mood/stress trends 01 / 07

Demographics — gender, age, grade & country 02 / 07

Lifestyle, support & social status 03 / 07

AI usage timing across the week 04 / 07

AI tool breakdown & stress/mood distribution 05 / 07

Deviation analysis — the Wednesday finding 06 / 07

Habit correlation testing (sleep, screen time) 07 / 07

04 — Key Findings

What the Data Revealed

Critical Finding — AI Usage & Academic Pressure

+28% AI Usage on Wednesdays

2,997 sessions vs Monday's 2,347 — a 650-session midweek spike, staying elevated into Thursday.

AI usage followed a clear midweek spike, peaking on Wednesday and staying elevated into Thursday before dropping across the weekend. This mirrors real-world research: 1 in 8 adolescents now use AI chatbots for mental health advice (RAND, 2025). Worth noting: the direction isn't certain — academic pressure may be driving the AI use, or the AI use itself may be adding to the stress. Either way, the right response is education, not restriction.

Finding 01

Emotional stability is the surprise

Average mood stayed above the scale neutral midpoint (6.0/10) and stress stayed below it (4.04/10), regardless of lifestyle level or demographics. Emotional stability — not fragility — was the dominant pattern.

Finding 02

Lifestyle doesn't predict stress

At Risk and Healthy Lifestyle students showed a stress difference of just 0.01 points (4.07 vs 4.06). Individual habits — sleep, screen time, journaling, exercise, meditation — were each tested separately and also came back flat.

Finding 03

Just under half cross the stress line

48.7% of students — 487 of 1,000 — recorded above-average stress levels. In a real population, just under half of all teens experiencing elevated stress would still be a significant public health signal.

Finding 04

AI spikes at academic pressure points

AI usage peaked +28% on Wednesdays vs Mondays and stayed elevated into Thursday. Teens appear to lean on AI precisely when they feel most stretched academically — a behavioural signal worth monitoring, though whether pressure drives the usage or usage adds to the pressure isn't yet clear from this data.

Finding 05

Support doesn't move the needle either

Average stress across support groups ranged only 0.01 points — and follow-up testing at the individual level confirmed no hidden effect either. Like lifestyle, feeling supported didn't separate students by outcome in this dataset.

Finding 06

Mood is above neutral across all groups

Across every lifestyle group, gender and country, average mood stayed between 5.93 and 6.06 — consistently above the neutral midpoint of 5.0. The dominant signal is moderate-to-positive emotional wellbeing, not crisis.

05 — Recommendations

What Schools, Parents & Policymakers Should Consider

Know their AI patterns and provide support

AI usage peaks on Wednesdays and stays elevated into Thursday. Learn which days a student leans on AI most, and show up for them then — not after.

Promote healthy routines — not screen restrictions

Students with healthy sleep still showed strong AI usage. The focus should shift from restricting AI to building intentional habits around it — sleep boundaries, study breaks, and purposeful use.

Teach AI literacy — don't just limit access

Guide students to use AI as a tool that enhances their thinking, not replaces it. Structured conversations about verifying AI outputs and trusting their own reasoning are essential.

Build their own stress toolkit directly

1 in 8 teens already use AI for mental health advice (RAND, 2025). Help students manage stress and mood on their own terms, so AI stays a support — not their only outlet.