Most students think publishing research is something that happens in graduate school. These high schoolers proved otherwise.
Below are examples of research by YRI Fellowship students, published or accepted at IEEE conferences or placed at science fairs. Each student worked 1:1 with a PhD mentor.
AI & Machine Learning Papers
Anomaly Detection in Cloud Systems
Student: Aditya Singla, 11th Grade, Cupertino High School Published: IEEE AAIML 2026 (Tokyo, Japan) Award: Best Oral Presentation
Aditya developed a novel multistep neural network approach to detect anomalies in online operational systems. His model processed over 117,000 features from 39,365 data points of IBM Cloud telemetry and detected 16 of 25 anomalies against 6 of 25 in prior research, a 2.7x improvement. He presented his findings at an international IEEE conference in Tokyo and won Best Oral Presentation.
What made it work: A clearly defined problem (cloud system failures), a novel two-step methodology, and strong quantitative results that improved on existing benchmarks.
Exoplanet Habitability Index
Student: Ruthwik Dhama, 11th Grade, Enloe High School Published: IEEE CONMEDIA 2025 (8th International Conference on New Media Studies)
Ruthwik created METHI, an ensemble-based machine learning framework that scores exoplanet habitability. Rather than analyzing one planet at a time, his system can evaluate all known exoplanets simultaneously.
What made it work: An ambitious scope matched with a practical ML approach. The framework had clear real-world applications for astronomical research.
Lung Cancer Biomarker Discovery
Student: Aarnav Bhat, 11th Grade Accepted: IEEE ICITSIF 2026
Aarnav used machine learning on genomic data to identify biomarkers for lung cancer detection, achieving an AUC of 0.983, near-perfect discrimination between cancerous and non-cancerous samples.
What made it work: Medical relevance, strong statistical results, and a clear contribution to the early detection field.
Environmental Science Papers
Air Pollution and Respiratory Disease
Student: Avyay G., 9th Grade Award: 1st Place Science Fair, State Qualifier
Avyay integrated air pollution exposure data with genetic susceptibility to predict respiratory disease risk using AI models including survival analysis and SHAP interpretability.
What made it work: Interdisciplinary approach combining environmental data with genetics, plus strong AI methodology for a 9th grader.
Neuroscience & Healthcare Papers
Astrobiology-Inspired Alzheimer's Detection
Student: Ayaan Rustagi, 10th Grade, Rouse High School Published: IEEE Conference Award: 3rd Place Science Fair
Ayaan applied astronomy techniques like pulsar timing to detect Alzheimer's from brain scans, achieving 88.9% accuracy. This cross-domain approach was novel and caught judges' attention.
What made it work: Creative cross-domain thinking that bridged two seemingly unrelated fields.
Common Patterns in Successful Student Papers
Looking across all these examples, successful high school research papers share these traits:
- Clear, specific research question: not "I want to study AI" but "Can a multistep neural network detect cloud anomalies better than existing methods?"
- Novel methodology or application: applying an existing technique to a new domain, or combining methods in a new way
- Quantitative results: accuracy scores, AUC values, percentage improvements that reviewers can evaluate
- Real-world relevance: papers that solve actual problems get published faster
- Strong mentorship: every student above worked with a PhD mentor who guided their methodology
Annotated Examples: How to Write Each Section
Reading finished papers shows you the standard. Seeing why each section works shows you how to hit it. Most scientific papers follow the IMRaD format:
- Introduction: Why did you do this?
- Methods: How did you do it?
- Results: What did you find?
- and
- Discussion: What does it mean?
The annotated excerpts below show what strong writing looks like in each section. They are illustrative samples written for this guide, not taken from published papers, and bracketed citations like [Author, Year] mark where a paper would cite its own sources. For planning the structure before you write, see our research paper outline guide.
Abstract
The abstract summarizes your entire paper in 150-300 words and must stand alone without the full paper.
Example 1: Biomedical research
Background: Diabetic retinopathy causes distance vision impairment or blindness in about 3.9 million people worldwide, and early detection and timely treatment are crucial to avoid irreversible vision loss. Early detection significantly improves outcomes, yet current screening methods require specialized equipment and trained ophthalmologists. Methods: This study developed a deep learning model to detect diabetic retinopathy from smartphone-captured retinal images. A MobileNetV3 neural network was trained on 500 images from three diabetes clinics, with ground truth labels from board-certified ophthalmologists. Results: The model achieved an area under the ROC curve of 0.94, with 91.2% sensitivity and 87.3% specificity for detecting any retinopathy. Performance was consistent across patient demographics. Conclusion: Smartphone-based screening could enable widespread diabetic retinopathy detection in resource-limited settings.
Why it works: It follows a Background-Methods-Results-Conclusion structure, includes specific quantitative results, and states the significance clearly.
Example 2: Environmental science
Urban heat islands contribute to increased energy consumption, heat-related illness, and reduced air quality in cities. This study analyzed the relationship between green space coverage and surface temperature across 50 locations in Phoenix, Arizona during summer 2024. Using Landsat 8 thermal imagery and NDVI vegetation indices, we found a strong negative correlation (r = -0.73, p < 0.001) between vegetation coverage and surface temperature. Locations with greater than 30% green coverage were on average 4.2°C cooler than locations with less than 10% coverage. These findings support expanding urban tree canopy as a strategy for heat mitigation, with implications for public health and urban planning.
Why it works: A clear problem statement, a specific methodology, key statistical results, and practical implications, all in one concise paragraph.
Introduction
The introduction establishes context, identifies the gap, and states your research question.
[Hook - Significance] Social media is now a daily habit for teenagers, with the average American teen spending 4.8 hours a day on social media platforms (Gallup, 2023). At the same time, 40% of US high school students reported persistent feelings of sadness or hopelessness in 2023 (CDC, 2023).
[Background - What's Known] Previous research has established correlations between social media use and mental health outcomes, including a small but significant relationship with anxiety symptoms [Author, Year]. However, the nature of this relationship remains debated. Some researchers argue that social media causes anxiety through social comparison and fear of missing out [Author, Year], while others suggest that anxious individuals may simply use social media more as a coping mechanism [Author, Year].
[Gap - What's Missing] Notably, most studies have examined total screen time without distinguishing between different types of social media use. Active use (posting, commenting, messaging) and passive use (scrolling, viewing) may have distinct psychological effects, yet few studies have examined this distinction in adolescent populations.
[Research Question/Hypothesis] The present study investigates whether active versus passive social media use differentially predicts anxiety symptoms in high school students. We hypothesize that passive use will show a stronger association with anxiety than active use, as passive consumption facilitates social comparison without the benefits of social connection.
| Section | Purpose | Words |
|---|---|---|
| Hook | Grab attention, show significance | ~50 |
| Background | What's known from prior research | ~100 |
| Gap | What's missing or unclear | ~50 |
| Research question | What you're investigating | ~50 |
Why it works: It starts with real-world significance, cites specific prior research, clearly identifies the gap, and moves from broad to specific before stating a testable hypothesis.
Methods
The methods section describes exactly what you did, in enough detail for replication.
Participants
Participants were 127 students (68 female, 59 male, mean age = 16.2 years, SD = 0.9) recruited from three high schools in suburban Illinois during Fall 2024. Inclusion criteria required participants to be aged 14-18, own a smartphone, and use at least one social media platform daily. Students with diagnosed anxiety disorders were excluded to focus on subclinical symptom variation. All participants provided informed assent, and parental consent was obtained. The study was approved by the Springfield High School IRB (#2024-031).
Measures
Social Media Use. Participants installed the Screen Time Tracker app on their smartphones for two weeks. The app automatically recorded daily usage of each social media application and categorized behaviors as active (posting, commenting, messaging) or passive (scrolling feed, viewing stories, watching videos) based on in-app activity patterns.
Anxiety Symptoms. The Generalized Anxiety Disorder 7-item scale (GAD-7; Spitzer et al., 2006) was administered at baseline and after the two-week monitoring period. The GAD-7 is a validated self-report measure with scores ranging from 0-21 (α = 0.89 in the present sample).
Procedure
Participants completed baseline surveys during school hours. They then installed the monitoring app with researcher assistance and used their phones normally for 14 days. Daily automated prompts reminded participants to keep the app running. At the end of the monitoring period, participants completed the follow-up GAD-7.
Data Analysis
Hierarchical multiple regression examined whether active and passive social media use predicted anxiety symptoms, controlling for baseline anxiety, age, and gender. All analyses were conducted in R version 4.2.1 using the stats and car packages.
Why it works: Specific participant demographics, clear inclusion and exclusion criteria, ethical approvals, measures described with citations, reliability reported, a chronological procedure, and the statistical approach stated.
Results
The results section presents findings objectively, without interpretation.
Descriptive Statistics
Participants used social media an average of 4.3 hours per day (SD = 1.8), with 1.2 hours (SD = 0.7) classified as active use and 3.1 hours (SD = 1.4) as passive use. Mean baseline GAD-7 score was 6.8 (SD = 4.2), and mean follow-up score was 7.1 (SD = 4.5). Descriptive statistics are presented in Table 1.
Primary Analyses
Hierarchical regression results are summarized in Table 2. In Step 1, baseline anxiety, age, and gender accounted for 62% of variance in follow-up anxiety scores, F(3, 123) = 67.4, p < .001. In Step 2, adding total social media use did not significantly improve prediction, ΔR² = .01, p = .18.
In Step 3, when active and passive use were entered separately, the model significantly improved, ΔR² = .04, p = .008. Passive social media use was a significant positive predictor of follow-up anxiety (β = 0.19, p = .003), while active use was not significant (β = -0.05, p = .42).
Figure 1 displays the relationship between passive use and anxiety symptoms, with higher passive use associated with greater anxiety.
Secondary Analyses
Exploratory analyses examined potential gender differences. The interaction between gender and passive use was not significant (p = .31), suggesting the relationship between passive use and anxiety was similar for male and female participants.
Why it works: Descriptive statistics come first, tables and figures are referenced, effect sizes and exact p-values are reported, and there is no interpretation, just facts.
How to report statistics:
| Statistic | Format | Example |
|---|---|---|
| Mean and SD | M = X.XX, SD = X.XX | M = 4.32, SD = 1.78 |
| Correlation | r = .XX, p = .XXX | r = .45, p = .003 |
| t-test | t(df) = X.XX, p = .XXX | t(125) = 2.34, p = .021 |
| ANOVA | F(df1, df2) = X.XX, p = .XXX | F(2, 124) = 5.67, p = .004 |
| Regression | β = X.XX, p = .XXX | β = 0.23, p = .012 |
| Chi-square | χ²(df) = X.XX, p = .XXX | χ²(2) = 8.45, p = .015 |
Discussion
The discussion interprets results and places them in context.
[Summary of Findings] This study examined whether active and passive social media use differentially predict anxiety symptoms in high school students. Results supported our hypothesis: passive social media use significantly predicted higher anxiety, while active use did not. These findings held after controlling for baseline anxiety, suggesting that passive use may contribute to anxiety symptoms rather than simply reflecting pre-existing tendencies.
[Interpretation] The differential effects of active versus passive use align with theoretical models emphasizing social comparison as a mechanism linking social media to mental health. Passive scrolling exposes users to curated, idealized content without the social connection benefits of interaction [Author, Year]. In contrast, active use, messaging friends, posting content, commenting on others' posts, may fulfill belonging needs that buffer against anxiety.
[Comparison to Prior Research] Our findings extend previous work showing overall correlations between social media use and anxiety [Author, Year] by identifying passive use as the key driver. This may explain mixed findings in prior studies that did not distinguish use types. Our effect size (β = 0.19) is in line with experimental work finding that limiting social media use improved well-being [Author, Year].
[Implications] These results have practical implications for adolescents, parents, and educators. Rather than limiting total screen time, interventions might more effectively target passive consumption specifically. Features that interrupt extended scrolling or encourage active engagement could be beneficial.
[Limitations] Several limitations should be noted. First, despite controlling for baseline anxiety, our correlational design cannot establish causation. Experimental manipulation of use patterns is needed. Second, our sample was predominantly suburban and may not generalize to urban or rural populations. Third, the two-week monitoring period may not capture longer-term patterns.
[Future Directions] Future research should experimentally manipulate passive versus active use to establish causality. Longitudinal studies could examine whether early passive use predicts anxiety development over years. Additionally, examining specific platforms and content types may reveal more nuanced patterns.
[Conclusion] In summary, this study provides evidence that not all social media use is equivalent for mental health. Passive consumption, but not active engagement, predicted anxiety symptoms in high school students. These findings point toward targeted interventions focusing on reducing passive scrolling rather than blanket screen time restrictions.
Why it works: It opens with a clear summary, interprets results with reference to theory, compares to prior research specifically, discusses practical implications, acknowledges limitations honestly, and ends with a strong conclusion.
Section Length Guide
| Section | Typical % of Paper | Example (10-page paper) |
|---|---|---|
| Abstract | N/A (separate) | 150-300 words |
| Introduction | 15-20% | 1.5-2 pages |
| Methods | 20-25% | 2-2.5 pages |
| Results | 20-25% | 2-2.5 pages |
| Discussion | 25-30% | 2.5-3 pages |
| References | N/A (separate) | As needed |
The most common mistake is vague writing. Instead of "there was a significant difference," write "Group A scored 23% higher than Group B (t(48) = 3.45, p = .001)." Use these examples for structure and style, but always write your own content. Never copy text.
How to Write Your Own
Most of these students started with no research experience. The difference was structured mentorship that gave them a topic, methodology, and writing support from week one.
The YRI Fellowship pairs you 1:1 with a PhD mentor matched to your interests for 10 weeks and guides you from research question to a first-author paper submitted to IEEE, Springer or Elsevier venues, with mentorship continuing at no extra cost until the paper is published. 87% of Fellows publish.
Where to go next
If this answered your question, the next ones are usually: journals that publish high school research, the best research programs for high school students, and how to write your first research paper. For formatting, see the research paper format guide and how to write a literature review.
Sources
- Gallup: Teens spend an average of 4.8 hours on social media per day (2023)
- CDC: 2023 Youth Risk Behavior Survey results (persistent sadness or hopelessness at 40%)
- World Health Organization: blindness and vision impairment fact sheet (3.9 million with diabetic retinopathy)
- JAMA Internal Medicine: Spitzer et al., 2006, a brief measure for assessing generalized anxiety disorder (GAD-7)





