HealthPredictor

Inspiration

People rarely recognise health problems early on, but they frequently begin with minor lifestyle choices like poor sleep, excessive screen time, and inactivity. I wanted to create a tool that takes commonplace behaviours and provides an understandable, AI-powered health risk estimate. It should be quick, easy, and meaningful. The objective was to raise awareness of wellness without requiring wearable technology or medical equipment.

How we built it

Using a structured lifestyle-wellness dataset that included sleep duration, water consumption, steps, BMI, calories, heart rate, and other information, I trained a Random Forest Classifier. During dataset exploration, I created derived features like daily screen-time impact and mood score to improve prediction quality.

With user input, the Streamlit app executes the entire prediction pipeline (no retraining required) after loading the trained model from a.pkl file. Clean sliders and numeric fields make up the simple, fast, and beginner-friendly user interface.

Challenges we ran into

Managing a small dataset while maintaining model performance

Creating derived features that benefit the model rather than detract from it

Keeping predictions interpretable while avoiding overfitting

Making the user interface simpler to make it more approachable and intuitive

completing a polished demo video in the allotted three minutes

What we learned

How to preprocess real-world lifestyle data

Building and tuning tree-based ML models

Creating derived features that meaningfully improve classifier performance

Deploying interactive web apps with Streamlit

Making a clean, simple UI that works for non-technical users

What HealthPredictor does

Takes lifestyle inputs (sleep, steps, calories, water, etc.)

Includes derived features like screen time and mood score

Runs them through a trained ML model

Returns a simple Low / Medium / High health-risk score instantly

It’s built to be fast, helpful, and beginner-friendly.

Built With

• numpy
• pandas
• pickle
• python
• scikit-learn
• streamlit