Targeting What Matters Early: Using Machine Learning to Identify Malleable Predictors of First-Term GPA

• In Event: Drivers of Educational Attainment

Abstract

First-term academic performance is a critical milestone that shapes long-term outcomes for college students, making the early college experience a key point for timely, effective intervention. While student wellbeing is widely acknowledged as integral to engagement, motivation, and academic success (Boulton et al., 2019; Korhonen et al., 2014; Lansford et al., 2017), it is rarely centered as a quantitative predictor of early academic performance. At the same time, there is growing interest in understanding how multiple dimensions of student identity and experience interact to influence early academic trajectories (Keller et al., 2022; Museus & Griffin, 2011).

This study uses random forests to examine how wellbeing, student characteristics, and academic preparation predict first-term GPA among five cohorts of incoming undergraduates (2018-2019 to 2022-2023) at a large public university in Oregon. Random forests offer a flexible, interpretable framework for identifying key predictors in high-dimensional data, especially where non-linear relationships and multicollinearity may obscure insights from traditional regression models (Breiman, 2001). The method also enables exploration of subgroup-specific patterns, revealing where interventions can be most impactful.

Findings show that alongside academic variables such as credits attempted, three malleable psychological constructs—belonging uncertainty, ability uncertainty, and self-assurance—emerged as among the most important predictors of first-term GPA. Of these, belonging uncertainty emerged as the most important wellbeing subconstruct: higher belonging uncertainty meant that students felt greater global concern about their social ties at the university, anticipating poor social fit (Walton & Cohen, 2007). Findings indicate that evidence-based programs and policies, such as an intervention that simply frames “social adversity as common and transient” (Walton & Cohen, 2011) may be effective in bolstering social and intellectual fit — and subsequently their academic performance.

By identifying which early psychological experiences matter most for academic success, this research supports the development of actionable policy solutions. It supports cross-sector collaboration among higher education leaders, researchers, and policymakers to design inclusive, data-informed interventions that support all students from the outset of their college journey – aligning closely with APPAM’s 2025 conference theme of collaboration, forging partnerships for transformative and resilient change. Full results will be shared in advance of the conference on GitHub (https://github.com/anweshaguha/inequalities-of-success).

Author

• Anwesha Guha, University of Oregon
  Presenting Author