Machine Learning to Identify Important Predictors of School Science Literacy Achievement: Evidence from the United States

Many educational studies have examined the relationships between school science literacy achievement and various school-level factors, including school resources, teacher quality, and class size. The primary purpose of the current study was to investigate the predictive ability of student-, teacher-, and principal-reported variables for school science literacy achievement using a machine learning approach. Using data from 238 US schools collected by the 2015 Program for International Student Assessment, we examined the relative importance of 48 predictors in predicting school science literacy achievement. The machine learning algorithm identified ten relatively important predictors of school science literacy outcomes. Of the ten predictors, seven were reported by the students. We discussed the results concerning implications for school leaders in setting program priorities.