Exploring Student Test-Taking Behaviors in PIRLS 2021 Using Machine Learning Techniques

• In Event: AERA Roundtable Session Wednesday 12:40 pm Four Seasons Ballroom 2-3
  In Roundtable Session: Artificial Intelligence in Education: Innovations in Assessment, Literacy, and Performance Analysis (Table 5)

Abstract

This study analyzes Singapore's digital PIRLS 2021 student data using machine learning methods to address three research questions. First, we applied a Collaborative Filtering algorithm to impute missing values in a sparse dataset, achieving reliable results with a high correlation (0.92) and R-squared value (0.85). Second, we trained an autoencoder to extract significant features from students’ process data, capturing patterns of test-taking behavior. Third, we used K-means clustering to classify students into five distinct groups based on these features. The results indicate clear distinctions in reading achievement among clusters, demonstrating that students' reading achievement has a significant relationship with their test-taking behavior. This showcases the effectiveness of our methods for meaningful analysis and interpretation.

Authors

• Zhushan Mandy Li, Boston College

• Jihang Chen, Boston College