Identifying Struggles and Tutorial Impact in Online Education Logs

Clustering and regression analysis on RIIID education logs to identify student difficulties.

<View the interactive 3D KMeans plot!>

---

Abstract

The purpose of this material is to accurately make predictions on student performance to tailor their education to their ability. Riiid is the source company, an AI EdTech company based in South Korea. Enabling higher efficiency in education improves student outcomes, which in turn opens doors for social mobility. Meeting students with support for exactly where they're struggling is an efficient use of both the student and educators time. This is important to educators, students, parents, and nearly everyone with a stake in education.

K-Means Clustering as well as a regularized least squares model was implemented. The k-means clustering revealed performance groups by capturing variation in accuracy, standard deviation, and number of attempts. The regularized least squares yielded with $\lambda=0.01$ achieved ~80.7% accuracy and 100% precision in predicting student success. Feature engineering was required on the original data to include metrics such as mean accuracy, consistency trends, and effectiveness to train the model. These findings suggest that predictive modeling with performance-based clustering can support the development of personalized education strategies, since educators can reliably determine whether a student needs help or falls within a performance group that is classified as intervention needing.

Methodology

---

Analaysis was implemented entirely within Julia. Libraries included were DataFrames.jl, Statistics.jl, LinearAlgebra.jl, Clustering.jl, CSV.jl, Random.jl, and PlotlyJS.jl for visualization.