Phone-Based Indoor Air Quality Mapping

ISEF Category: Systems Software

Subcategory: Mobile Apps  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

Most classrooms never measure the air you breathe, even when the room feels stuffy. Your phone already carries sensors that can pick up clues. If you combine those clues with user labels and spot checks, you can build a map that points to bad air before anyone notices a headache. That makes this project useful, testable, and very real.

What Is It?

This project uses a phone as a small sensing station. A barometer tracks pressure changes, a microphone can pick up HVAC hum, and user labels can mark when a room feels crowded, stale, or well ventilated. You then train a model to infer a proxy for indoor air quality, such as likely CO2 buildup or particle-heavy conditions, instead of trying to read those gases directly from the phone.

Think of it like weather prediction for a room. You do not measure every air molecule. You look for patterns that often travel together. A room with weak ventilation, rising occupancy, and certain sound or pressure patterns may also show worse air quality on a real sensor. Your job is to test whether those patterns actually hold up.

Why This Is a Good Topic

This is a strong science fair topic because it mixes real-world need with a clear testable model. Schools, libraries, and homes all need better ventilation awareness, and phone data gives you a low-cost way to explore it. You can study sensor fusion, classification, calibration, and validation, which gives you room to do real computer science and data science. You can also make the project stronger by checking whether your proxy works across different rooms, times of day, and building types.

Research Questions

• How does adding microphone-based HVAC features change the accuracy of an indoor air quality proxy model?
• What is the effect of including barometer trends on predicting spot-checked CO2 levels?
• Does a model trained on user-labeled room conditions generalize to new classrooms or bedrooms?
• To what extent can phone sensor data predict when a room crosses a poor ventilation threshold?
• Which feature set, barometer only, microphone only, or both together, best matches SCD30 spot checks?
• How does building type affect the error rate of a community-sourced indoor air quality map?
• To what extent do occupancy labels improve the proxy model compared with sensor data alone?

Basic Materials

• Smartphone with barometer and microphone access
• Second smartphone for testing or comparison
• Laptop or desktop computer
• Spreadsheet software
• Free data notebook or form app
• Notebook for room labels and observations
• Access to multiple indoor spaces
• Quiet reference room for baseline measurements
• Tripod or phone stand
• Optional external thermometer for context

Advanced Materials

• Smartphones with different sensor models
• SCD30 sensor or equivalent CO2 reference sensor
• Raspberry Pi or laptop for data collection, if needed
• Sound level meter or calibrated microphone setup
• Portable particulate matter sensor for spot checks
• Wi-Fi router or BLE logger for occupancy context
• Reference thermometer and hygrometer
• Data storage pipeline for time-stamped sensor fusion
• Python environment for modeling
• Statistical software for validation and error analysis

Software & Tools

• Python: Cleans sensor logs, builds features, and trains simple machine learning models.
• Pandas: Organizes time-stamped phone and label data into analysis tables.
• Scikit-learn: Trains and tests baseline classifiers or regressors for air quality proxies.
• ImageJ: Helps if you want to analyze any visual room markers or plots from screenshots.
• Google Forms: Collects user labels about room crowding, comfort, and perceived ventilation.

Experiment Steps

1. Define the air quality outcome you want to predict, such as a CO2 proxy, a ventilation class, or a comfort label.
2. Choose the phone signals you will compare, then decide whether you will test them alone or in combination.
3. Design a labeling plan that records room use, occupancy, and any reference sensor checks at the same time.
4. Build a calibration set that links phone features to spot-checked ground truth from the reference sensor.
5. Test whether your model still works in new rooms, new buildings, or new times of day.
6. Plan error analysis so you can see where the proxy fails and which sensor features cause the failure.

Common Pitfalls

• Collecting phone data without strict time stamps, which breaks alignment with the reference sensor.
• Using only one room type, which makes the model look good in one place and fail everywhere else.
• Letting user labels stay vague, which creates noisy targets like "bad air" with no measurable meaning.
• Ignoring sensor differences between phone models, which can make barometer and microphone features drift across devices.
• Skipping validation against a real air-quality reference, which leaves you with a pattern detector instead of a verified proxy.

What Makes This Competitive

A competitive version of this project goes beyond a simple app demo. You would compare multiple sensor combinations, test on new buildings, and report real error metrics, not just accuracy. Strong entries also look for failure modes, like whether the proxy breaks in noisy halls, crowded rooms, or buildings with weak HVAC systems. If you can show when the map works, when it fails, and why, your project starts to look like research.

Project Variations

• Use only the barometer and occupancy labels to see how far pressure trends alone can predict poor ventilation.
• Swap the HVAC hum feature for a calibrated sound spectrum feature and test whether frequency bands improve the proxy.
• Change the outcome from CO2 proxy to a room comfort score and compare how human labels match sensor fusion results.

Learn More

• PubMed: Search review articles on indoor air quality, ventilation, and CO2 as a proxy for occupancy.
• NIH PubMed Central: Find open-access papers on mobile sensing, sensor fusion, and indoor environmental quality.
• NASA Earthdata: Explore articles and data tools on atmospheric sensing and signal calibration methods.
• NOAA Air Resources Laboratory: Read background material on air chemistry, aerosols, and measurement concepts.
• MIT OpenCourseWare: Search courses on machine learning, signal processing, and data analysis for sensor projects.
• USGS Water Science School: Use the data literacy sections to practice time series, calibration, and field measurement thinking.