👋 Hi, I’m Sylvia. I’m a graduate student in Education Policy Analysis at the Harvard Graduate School of Education. With a background in economics and a growing skill set in data science, I’m exploring how we can use data not just to describe education systems, but to challenge assumptions, ask sharper questions, and shape better decisions.
I bring a unique perspective to education-focused data analytics in using data to investigate problems I care about: student well-being, equity in school funding, digital learning, and school climate. Each project here represents a new learning curve for me, from deep learning and predictive modeling to scraping enrollment patterns and building policy-relevant dashboards.
This portfolio is a snapshot of what I’ve built so far. But more importantly, it’s a reflection of how I think, what I value, and how I approach complex questions with empathy and structure. I’m actively seeking data analyst roles in education research, EdTech, or nonprofit analytics, where I can grow while contributing to meaningful work.
Can AI prevent course dropouts by predicting student struggles 4 weeks in advance? This project answers with a resounding YES. I built this project to explore how AI can proactively support students before they fall behind. I simulated Canvas LMS (Learning Management System) data, then developed an LSTM-based predictive model to identify at-risk students weeks in advance. I then built a recommendation system that delivers personalized, actionable feedback, helping students, instructors, and advisors intervene early and improve course outcomes. Finally, I visulized the simulated dataset into a [Tableau Dashboard] (https://public.tableau.com/app/profile/leyan.li/viz/CanvasRecommendationSystem/CourseandStudentPerformanceOverview)for the course and student performance distribution overview.
Business Impact: Proactive intervention system to reduce student dropouts and improve academic performance
Dataset: Simulated Canvas LMS data with 2,000 students, 16 weeks of engagement, 40,000 labeled examples
Skills: Python, Machine Learning, Recommendation Systems, Data Simulation, Educational Analytics
Using survey data from Harvard’s Making Caring Common initiative, I developed a risk classification system to identify students vulnerable to bullying. Key features included perceptions of safety, belonging, and discrimination, helping educators direct support where it’s needed most.
Business Impact: Early-warning system to support student safety and school climate initiatives
Dataset: 8,366 student responses, 195 features
Skills: R, Machine Learning
This project investigates how per-pupil spending and district-level SES interact to shape academic outcomes. I used multilevel modeling to explore whether spending impacts are more pronounced in lower-SES districts, providing insights relevant for policy advocates and education finance reform.
Business Impact: Informs equitable funding formulas and policy debates on school finance
Dataset: SEDA (Stanford), CCD (NCES); 2015–2019
Skills: R (lme4, dplyr, ggplot2), data wrangling, multilevel modelling (random intercept)
I explored what separates students who finish MOOCs from those who drop out by testing six machine learning models. Random Forest performed best, identifying certificate earners with an F1 score of 0.737. This work reflects my interest in understanding learner behavior in online environments.
Business Impact: Supports retention and design strategy for online learning platforms
Dataset: HarvardX/MITx anonymized learner data (~100k+ records)
Skills: Python (scikit-learn, matplotlib, seaborn, xgboost), R, machine learning
This web scraping and visualization project traces how school demographics shifted in one Massachusetts city known for its school choice policies. Inspired by a case I studied in class, I analyzed patterns in diversity, equity, and parental choice over 20 years. I organized the data scraped into a Tableau dashboard for interactive visualizations.
Business Impact: Helps districts evaluate choice policies and equity outcomes
Dataset: Somerville School District enrollment reports (web scraped)
Skills: R (rvest, xml2, tidyverse), data visualizations
Using deep learning and facial recognition, this project aims to support educators in understanding students’ emotional states in digital classrooms—where non-verbal cues are often lost. By building and evaluating three convolutional models, I explored how real-time emotion detection could power early mental health alerts or responsive teaching tools.
Business Impact: Potential to integrate into EdTech platforms to support student well-being and engagement
Dataset: 28×28 pixel grayscale images, ~29,000 samples across 4 emotion classes
Skills: Python (TensorFlow, Keras, Numpy, Pandas), Deep Learning
Email: sylvialileyan@gmail.com
LinkedIn: linkedin.com/in/sylvia-leyan-li