Mental Health Signal: Predicting Student Wellness from Daily Habits

A machine learning project by Sandeep Kumar

Show Image Replace this with a screenshot of your app's input form or results screen.

Why I Built This

College life runs on habits most students never stop to measure — screen time, sleep, study hours, stress. I wanted to know: could those everyday numbers predict how a student is really doing?

That question became Mental Health Signal, a Streamlit app that takes a student's daily routine and turns it into a simple 0–10 wellness score, powered by a machine learning model trained on lifestyle and academic data.

Note: This is an informational analytics tool, not a diagnostic or clinical instrument. It's designed to spot patterns, not replace professional advice.

The Problem

Most students never connect the dots between their daily habits and how they feel. Symptoms of burnout — poor sleep, low activity, high stress — often build up quietly before anyone notices. I wanted a lightweight tool that could:

Take a handful of everyday inputs (no medical data required)
Instantly generate a wellness "signal" score
Make the connection between habits and wellbeing visible and easy to understand

How It Works

The app collects information across three categories, then feeds them into a trained ML model to generate a score.

1. Profile

Age
Gender
Country

2. Academic & Digital Habits

Academic level (High School / Undergraduate / Graduate)
Most-used social platform
Primary purpose of use (Networking, Education, Entertainment, News)
Average daily screen time
Daily phone unlocks

3. Lifestyle & Stress

Study hours per day
Physical activity per day
Sleep per night
Perceived stress level (Low → Very High)

Once submitted, the model processes these inputs and returns a predicted wellness score from 0–10, giving the user a quick, data-driven read on their current routine.

Tech Stack

LayerTools UsedLanguagePythonData HandlingPandas, NumPyModelingScikit-learnVisualizationMatplotlib, SeabornFrontend/InterfaceStreamlitDeploymentRender

Building the Model

The core of the project was turning raw lifestyle data into something predictive. The main steps:

Data preprocessing — cleaning and structuring habit-based inputs (screen time, sleep, stress, etc.) into a usable feature set.
Feature engineering — converting categorical inputs (like stress level or platform choice) into model-ready numeric features.
Model training — using Scikit-learn to train a regression-based model that maps habits to a wellness score.
Evaluation — testing the model's predictions against held-out data to check consistency.
Deployment — wrapping the trained model in a Streamlit interface and deploying it on Render for public access.

Key Features

Simple, form-based UI — no technical knowledge needed to use it
Instant scoring — real-time prediction after form submission
Human-readable output — a score, not raw model internals
Clear framing — explicitly labeled as informational, not clinical, with a note encouraging users to seek real support if they're struggling

What I Learned

Feature engineering matters more than model complexity. Cleaning and structuring lifestyle data properly had a bigger impact on output quality than swapping algorithms.
UX matters for ML tools. A model is only useful if someone can actually interact with it — building the Streamlit form taught me a lot about translating technical inputs into plain-language questions.
Framing sensitive topics responsibly. Since this touches on mental health, I made sure the app is clearly labeled as an informational tool, not a diagnostic one — a small but important design decision.

Try It Yourself

🔗 Live App: mental-health-score-1-5ltv.onrender.com 🔗 GitHub: github.com/sandeepkumar7860

What's Next

Some directions I'd like to explore:

Adding a trend view so users can track their score over time
Expanding the feature set (e.g., diet, social interaction frequency)
Comparing multiple model types (Random Forest, XGBoost) for better accuracy

Have feedback or ideas for this project? Reach out on LinkedIn or check out more of my work on GitHub.