Attention-Aware Homework Assistant for ADHD Study Sessions

ISEF Category: Robotics and Intelligent Machines

Subcategory: Cognitive Systems  ·  Difficulty: Advanced  ·  Setup: Home Setup  ·  Time: 1 to 2 Months

The Hook

A timer can say you are working, even when your brain has checked out. That is the gap this project targets. You build a system that watches simple signals, then decides when a tiny nudge might help. The real test is whether it beats a plain Pomodoro timer.

What Is It?

This project builds a study assistant that estimates whether you are on task during homework. It uses two simple signals from a webcam and a microphone, gaze direction and ambient sound entropy. Gaze direction means where your eyes and face seem to point. Sound entropy means how varied or messy the background noise is. Think of it like a coach that looks at both your body position and the room around you before deciding whether you need a reminder.

The system turns those signals into an on-task probability. A probability is just a number from zero to one that acts like a confidence score. If the score drops, the assistant sends a minimal nudge, such as a visual cue or a soft prompt. You compare that design with a Pomodoro-only control, which gives reminders on a fixed schedule no matter what you are doing. That comparison helps you test whether adaptive nudges work better than a one-size-fits-all timer.

Why This Is a Good Topic

This is a strong science fair topic because you can test a clear claim, whether adaptive nudges keep a student engaged better than fixed-timer reminders. You can measure real outcomes such as task completion, time off task, response to nudges, and false alarms. The project connects to ADHD support, study habits, and assistive technology, which gives it real-world meaning. You can also learn machine learning basics, signal processing, experimental design, and statistical comparison in one project.

Research Questions

• How does combining gaze direction and ambient sound entropy change on-task probability estimates compared with gaze alone?
• What is the effect of adaptive nudges on task completion compared with a Pomodoro-only timer?
• Does a webcam-based attention model reduce off-task time during homework sessions?
• To what extent do sound-based features improve detection of distraction in a quiet room versus a noisy room?
• Which nudge style leads to fewer interruptions, a visual cue, or a short text prompt?
• How does model accuracy change when you train on one student and test on a different student?
• What is the effect of different decision thresholds on false alarms and missed distractions?

Basic Materials

• Laptop or desktop computer with webcam and microphone.
• Quiet study space with consistent lighting.
• External desk lamp for stable face lighting.
• Headphones for the participant, if audio feedback is tested.
• Google Sheets or Excel for logging session data.
• Notebook or form for recording task completion and self-report notes.
• A simple Pomodoro timer app for the control condition.
• Python installed on a personal computer.
• OpenCV and MediaPipe for webcam feature extraction.
• A distraction-free reading or worksheet task with answer keys for scoring.

Advanced Materials

• Laptop or desktop computer with webcam and microphone.
• External USB webcam with adjustable placement.
• External microphone for cleaner ambient sound capture.
• Second monitor or tablet for logging experimental events.
• Python environment with scikit-learn, NumPy, pandas, and Matplotlib.
• PyTorch or TensorFlow if you test a deeper model.
• OpenFace or MediaPipe for facial landmark and gaze-related features.
• Audacity or similar audio tool for inspecting noise profiles.
• Secure spreadsheet or database for session metadata.
• IRB-style consent and assent forms, if your school requires human-subject review.

Software & Tools

• Python: Runs your signal extraction, model training, and analysis scripts.
• OpenCV: Captures webcam frames and helps detect face position and motion.
• MediaPipe: Tracks face landmarks that can support gaze-related features.
• scikit-learn: Builds and compares simple classifiers and evaluation metrics.
• ImageJ: Lets you inspect image brightness and frame consistency if you test lighting effects.

Experiment Steps

1. Define the attention signal you will predict, such as on-task versus off-task, and write a clear rule for labeling sessions.
2. Choose one baseline control, then decide how your adaptive system will differ from a fixed Pomodoro timer.
3. Plan the webcam and audio features you will extract, then separate train, validation, and test sessions before you collect data.
4. Build a small calibration set so you can turn raw gaze and sound features into a usable score.
5. Design outcome measures that matter, such as task completion, false nudges, and missed distractions.
6. Preplan the comparison test you will use, then decide how you will summarize results across multiple sessions and participants.

Common Pitfalls

• Treating face direction as real attention, which can mislabel someone as on task when they are daydreaming.
• Collecting audio in a noisy room, which can make sound entropy reflect background chaos instead of distraction.
• Testing only one participant, which makes the model look better than it really is.
• Letting the adaptive system and the Pomodoro control use different task types, which confounds the comparison.
• Choosing a nudge threshold after seeing results, which inflates performance and weakens the experiment.

What Makes This Competitive

A competitive version of this project needs more than a working app. You would need a clean comparison, strong labels, and clear metrics for both accuracy and user impact. A tougher angle would test whether one signal helps only in certain rooms, tasks, or user states, then analyze those cases separately. You could also compare multiple models or threshold rules and show which trade-off best balances fewer false alarms with fewer missed distractions.

Project Variations

• Test the same attention model on reading comprehension, math homework, or puzzle solving instead of a generic study session.
• Replace the webcam gaze feature with head pose only, then compare how much accuracy you lose.
• Swap the ambient sound entropy feature for keyboard activity patterns, then test whether the combined model improves more or less.

Learn More

• NIH PubMed: Search for review articles on ADHD, attention monitoring, and digital interventions to ground your project in medical research.
• NASA Open Courseware and university machine learning lectures: Look for free courses on classification, feature selection, and evaluation metrics.
• OpenCV documentation: Find examples for webcam capture, face detection, and frame processing.
• MediaPipe documentation: Use the face landmark tools to estimate head pose and related visual features.
• NOAA or government audio and signal resources: Search for introductory material on entropy, noise, and signal variation if you need a plain-language refresher.
• Peer-reviewed journals such as Computers in Human Behavior and Journal of Attention Disorders: Search PubMed, Google Scholar, or your school library for studies on digital nudges, attention support, and ADHD.