Neural Network Olfactory Memory Richness

ISEF Category: Behavioral and Social Sciences

Subcategory: Behavioral Neuroscience  ·  Difficulty: Intermediate  ·  Setup: Home Setup  ·  Time: 1 to 2 Months

The Hook

A single smell can pull up a childhood scene faster than a photo. That makes odor a strong clue for memory. You can turn that effect into a dataset, then test whether a neural network can predict which scents trigger richer autobiographical memories. You do not need a university lab to start.

What Is It?

Olfactory-evoked autobiographical memory richness means how vivid, detailed, and emotional a memory feels after you smell something. A scent is like a shortcut key in a computer, it can unlock stored experiences with very little warning.

A neural network is a pattern-finding model that learns from many examples. In this project, it can look at scent labels, participant ratings, and memory scores, then learn which combinations are linked to richer memories. The goal is not to read minds, but to see whether the model can predict patterns better than a simple rule.

Why This Is a Good Topic

This is a good science fair topic because you can collect data with common household scents and a free Google Form, then turn the answers into a clean prediction problem. It connects to memory research, sensory perception, and everyday experiences, so the question feels real instead of abstract. You can learn survey design, data cleaning, and model evaluation without needing a wet lab.

Research Questions

• How does scent familiarity affect memory richness ratings?
• What is the effect of odor category, such as food, spice, or soap, on memory vividness scores?
• Does rated pleasantness predict richer autobiographical memories?
• To what extent can a neural network predict memory richness from scent features and survey ratings?
• Which input set gives better prediction accuracy, scent name alone, or scent name plus familiarity and pleasantness?
• How does participant age or grade level change memory richness ratings for the same scent?

Basic Materials

• Common household scent samples in opaque containers, such as vanilla, coffee, cinnamon, soap, and citrus.
• Clean cotton pads or unscented paper strips for carrying the scents.
• Labels, masking tape, and a marker.
• Phone or laptop with Google Forms and a spreadsheet.
• Notebook or lab log for participant codes and notes.
• Consent and assent forms for any school-approved data collection.

Advanced Materials

• Standardized odorant set with known concentrations.
• Olfactometer or scent delivery device.
• Quiet testing room with airflow control.
• IRB or school ethics approval paperwork.
• Python or R workstation for model training.
• Secure database or coded spreadsheet for storing responses.

Software & Tools

• Google Forms: Collects odor-memory ratings from many participants.
• Google Sheets: Organizes responses and helps you clean the dataset.
• Python: Trains the neural network and compares it with simpler baseline models.
• Jupyter Notebook: Keeps your analysis, charts, and notes in one place.
• scikit-learn: Splits the data, benchmarks models, and checks accuracy.

Experiment Steps

1. Define the outcome you will predict, such as vividness, detail, or emotional intensity, and keep the scale the same for every scent.
2. Choose a scent set and decide which features you will record for each one, such as familiarity, category, and pleasantness.
3. Design the survey flow so every participant sees the same instructions, answer choices, and scent order rules.
4. Build a baseline analysis first, then train the neural network and compare its performance against the simpler model.
5. Plan your holdout test, error checks, and group comparisons so you can explain when the model succeeds or fails.

Common Pitfalls

• Showing the scent name before the rating, which can inflate memory scores through expectation.
• Using scents with very different strengths, which makes intensity look like memory richness.
• Collecting too few ratings for each scent, which leaves the model with noisy labels.
• Changing the wording of the memory scale across sessions, which makes the responses hard to compare.
• Training and testing on the same participants, which makes the model look better than it really is.

What Makes This Competitive

A stronger version of this project does more than sort scents by score. It tests whether familiarity, pleasantness, intensity, or scent category actually drives memory richness, then compares a neural network with simpler models. A careful holdout split, participant-level analysis, and clear error reporting can make the project much stronger. That kind of design shows you understand both the behavior and the model.

Project Variations

• Test whether food, spice, and cleaning scents produce different memory richness scores.
• Swap the neural network for logistic regression or a random forest and compare prediction quality.
• Add a second analysis that checks whether familiarity or pleasantness changes the model's errors across participants.

Learn More

• PubMed: Search review articles on odor-evoked autobiographical memory, olfaction, and emotion.
• NCBI Bookshelf: Read free chapters on memory, sensation, and experimental design.
• Chemical Senses: Find peer-reviewed articles and reviews on smell perception and memory.
• MIT OpenCourseWare: Search for free lectures on introductory machine learning and model evaluation.
• OpenStax Psychology 2e: Review memory, perception, and research methods in a free textbook.