Machine Learning Insights into Multilingual Children's English Reading Achievement: Evaluating the Predictive Power of Kindergarten Factors on Elementary English Reading Achievement

Machine Learning Insights into Multilingual Children's English Reading Achievement: Evaluating the Predictive Power of Kindergarten Factors on Elementary English Reading Achievement

Existing machine learning research that examines factors influencing student’s academic achievement largely focuses on monolingual rather than multilingual students. To address this gap, we employ machine learning to analyze the Early Childhood Longitudinal Study, Kindergarten Class of 2010-11 (ECLS-K:2011) data of 3,542 nationally representative multilingual children from 970 U.S. schools, who were followed from kindergarten (Fall 2010) to 5th grade (Spring 2016). Our study models and compares the predictive power of six key factors in kindergarten (early reading ability, cognitive and language skills, socio-emotional skills, child characteristics, home literacy environment, and school/classroom characteristics) on English reading achievement across primary years. Using Random Forest and Elastic Net, we will predict English reading achievement from 1st to 5th grade based on the kindergarten predictors and analyze how the predictors’ importance evolves over time. Preliminary results indicate strong correlations within cognitive and language skills but weak correlations across other predictor factors. We are currently running machine learning analyses to answer our research questions. We hypothesize that initial English reading achievement will be the strongest predictor but diminishes over time, with other factors such as cognitive and language skills taking dominance in later elementary school. The findings of this study will suggest the significance of each key factor in multilingual children's long-term English reading achievement across primary years.