😉 StudyWELL — Student Habits & Mental Health Segmentation

A machine learning web application that analyzes student lifestyle habits and mental health patterns to automatically classify them into behavioral segments, then provides personalized AI-generated improvement plans.

Live Demo

Open StudyWELL

💡 Try the app to discover your student profile — answer 4 short sections about your daily habits and get a personalized AI improvement plan in seconds.
⚠️ If you're forking this project: don't forget to add your API key in Streamlit Cloud.
Go to Manage App (bottom right of your app) → Settings → Secrets → paste:

GROQ_API_KEY = "gsk_..."

Without this the AI chatbot won't work.

---

Project Overview

Universities treat all students the same way — one generic advice fits nobody. StudyWELL identifies which type of student you are based on your daily habits and mental health indicators, then gives you targeted, specific advice instead of generic recommendations.

How it works

1. Data — Two Kaggle datasets merged: student lifestyle habits (80,000+ students) + mental health indicators
2. ML Model — K-Means clustering discovers 4 natural student segments from 25 behavioral features
3. Validation — PCA scatter plot, radar chart, and box plot confirm segments are meaningful
4. App — Student answers 4 sections of simple questions → model classifies them → AI generates personalized plan

The 4 Student Segments

Segment	Profile
🔴 High-Anxiety Low-Motivation Students	High stress and anxiety, low motivation despite studying
🟢 High-Screen-Time Relaxed Students	High social media usage, low academic engagement
🟡 High-Screen-Time Burned-Out Students	Excessive screen time combined with academic burnout
🩵 Moderate-Study Low-Stress Students	Balanced habits, moderate performance

Project Structure

student-habits-mental-health-segmentation/
├── app/
│   └── app.py              ← Streamlit web application
├── models/
│   ├── kmeans.pkl          ← trained K-Means model
│   ├── scaler.pkl          ← fitted StandardScaler
│   ├── features.pkl        ← feature names list
│   └── cluster_names.pkl   ← cluster labels mapping
├── data/
│   └── processed/
│       └── students_clustered.csv
├── radar_clusters.html     ← interactive radar chart
├── pca_clusters.html       ← PCA scatter plot
├── boxplot_scores.html     ← exam score distribution
├── script.ipynb            ← full ML notebook
└── requirements.txt        ← all dependencies for deployment

ML Pipeline

Two CSV datasets (80,000+ students)
        ↓
Merge on student_id
        ↓
Clean + encode categoricals (LabelEncoder)
        ↓
Select 25 behavioral features
        ↓
Scale features (StandardScaler)
        ↓
Find optimal K (Elbow + Silhouette → K=4)
        ↓
Train K-Means clustering
        ↓
Name and interpret clusters
        ↓
Deploy with Streamlit + Groq AI chatbot

Datasets

Dataset	Source	Size	Features
Student Habits & Academic Performance	Kaggle	80,000 rows	31 columns
Student Mental Health & Burnout	Kaggle	150,000 rows	20 columns

Tech Stack

Layer	Technology
Language	Python 3.10
ML	scikit-learn (KMeans, StandardScaler, PCA)
Data	pandas, numpy
Visualization	plotly
App	Streamlit
AI Chatbot	Groq API (Llama 3.3 70B)
Deployment	Streamlit Cloud

Run Locally

# Clone the repo
git clone https://github.com/amyaby/student-habits-mental-health-segmentation.git
cd student-habits-mental-health-segmentation

# Create virtual environment
python -m venv venv
source venv/bin/activate  # on Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

Why requirements.txt? Streamlit Cloud starts from a fresh empty server — it has nothing installed. It reads requirements.txt and installs each library listed there before running your app. Your local machine already has everything in your venv, but the cloud knows nothing until you tell it. Without this file: no pandas, no sklearn, no groq, app crashes immediately.

# Add your Groq API key
mkdir .streamlit
echo 'GROQ_API_KEY = "gsk_..."' > .streamlit/secrets.toml

# Run the app
streamlit run app/app.py

Results

• K=4 clusters identified using Elbow method + Silhouette score
• Clusters are visually separated in PCA 2D projection
• Each cluster shows distinct behavioral fingerprint in radar chart
• Exam score distributions differ meaningfully across segments

Author

Amy Aby — Engineering student, ENSA Berrechid
S8 Data Mining Project