Individual, Family, and Community Education ETDs

Publication Date

Summer 7-28-2026

Abstract

High school dropout remains a persistent educational challenge with significant socio-economic consequences. This dissertation investigates dropout prediction through three interrelated studies utilizing machine learning (ML) and national longitudinal data. A systematic review study identified a surge in ML applications since 2015 and a critical need to move beyond student-level metrics to include broader contextual factors. Comparative analysis of ML methods demonstrated that while Linear Discriminant Analysis (LDA) exhibits the best performance at standard thresholds, lowering the classification threshold enhances sensitivity across all models, and informed the development of a Dropout Risk Framework encompassing academic, belief, behavioral, and contextual domains. Longitudinal comparison between the 2000s (ELS:2002) and 2010s (HSLS:09) cohorts found that while GPA and attendance remain foundational, socioemotional factors have gained predictive prominence. Ultimately, this dissertation affirms that calibrated ML models provide actionable insights for early warning systems.

Keywords

Machine Learning, High School, Dropout Prediction, At-risk students, Educational Data Mining, Socioemotional Factors

Document Type

Dissertation

Language

English

Degree Name

Educational Psychology

Level of Degree

Doctoral

Department Name

Individual, Family, and Community Education

First Committee Member (Chair)

Yu-Yu Hsiao

Second Committee Member

Carolyn J. Hushman

Third Committee Member

Yen Pham

Fourth Committee Member

Shengjie Lin

Available for download on Friday, July 28, 2028

Share

COinS