Identifying individual cognitive and motivational profiles predictive of academic growth: A combined machine learning and person-centered approach

Identifying malleable predictors of academic achievement is critical for supporting individual differences in learning outcomes and informing targeted interventions. However, practical constraints often require reducing the number of predictors while still accounting for meaningful variance. In this study, we combined two machine learning approaches (ridge regression and lasso regression) with a person-centered technique, latent profile transition analysis (LPTA), to isolate key cognitive and motivational factors that differentiate learners and predict academic growth. Using a large, nationally representative longitudinal dataset, machine learning analyses identified three robust predictors from 14 propensity variables: prior reading skills, motivation, and working memory. Subsequent LPTA revealed five distinct profiles of learners based on different combinations of these variables, with most children remaining in stable profiles across kindergarten and first grade, though some showed upward transitions. Importantly, these profiles transcended socioeconomic status and diagnostic categories, and they significantly predicted growth in mathematics achievement, a skill not used to create the profiles. Findings highlight meaningful and stable individual differences in cognitive and motivational profiles that shape learning trajectories, with implications for theory development, early identification, and the development of tailored intervention strategies. • We used machine learning (ridge/lasso regression) to identify a small set of malleable predictors of early academic achievement. • Five distinct learner profiles based on reading, working memory, and motivation emerged. • Profiles showed considerable stability across kindergarten and first grade and cut across SES and diagnostic categories. • Profiles predicted longitudinal growth in academic outcomes. • Findings highlight the value of integrating variable-centered and person-centered methods to capture individual differences in learning.