ÌÇÐÄvlogÃÛÌÒ

"L'art pour I'art"?: The Infused-Acquired Duplex Ordo of Art as an Intellectual Virtue

This dissertation, utilizing a Thomistic analysis, investigates the intellectual virtue of ars (art) in relation to the coexistence of the infused and acquired virtues within the cognitive structure of the Christian artist. In other words, this work analyzes the relationship between acquired ars and infused ars, with their distinct ends and interrelated configuration, as grounded in the perfection of human nature through the supernatural reality of grace. The basic formulation of this interrelationship is as follows: infused ars as directed to the production of sacred artwork builds upon acquired ars as directed to the production of secular artwork, but infused ars completes and fulfills acquired ars as meritoriously directing its proximate end to the human person’s supernatural end. As a whole, this project – by focusing on the intellectual virtue of ars – unites Thomistic virtue ethics and aesthetics in a heuristic synthesis.

This analysis is divided into six chapters. First, I articulate the historical crisis in art and aesthetics that has contributed to the internal alienation between the categories of artist, artisan, and beauty. Second, I organize a developed Thomistic aesthetic that unites St. Thomas Aquinas’s understanding of ars as an intellectual virtue proportional to the practical intellect, a careful reflection on the Thomistic characterization of beauty, and a synthesis of ars and beauty in light of the empirical fact of historical development in art. Third, I examine the unity within the Christian artist’s rationality by utilizing Alasdair MacIntyre’s conception of tradition-dependence and Bernard Lonergan’s theology of conversion. Fourth, I develop this synthesis by drawing attention to the external social/cultural conditions needed for ars to flourish, focusing on the dialectical structure of artwork and the internal movement between traditions in liturgical history. Fifth, I revisit the general debate concerning the coexistence of the acquired and infused virtues in the same Christian subject. Sixth, I synthesize the distinct dimensions of the infused-acquired duplex ordo of ars by illustrating the manner in which (i) infused ars builds upon preexisting acquired ars and (ii) acquired ars is directed by infused ars to a final supernatural end.

INTEGRATING AI AND EDUCATION DATA FOR PRIVACY-PRESERVING PREDICTION OF STUDENT SUCCESS

Student retention remains a persistent challenge in higher education; this dissertation addresses it through three interconnected studies that advance methods for predicting and supporting at-risk students while enabling privacy-preserving collaboration among institutions.

The first study surveys transformer-based text classification for educational applications across six dimensions: data modality, model size, input length, accuracy, computational cost, and safety (privacy, bias, and explainability). Although roughly 60 to 80 percent of organizational data pairs text with tabular fields, multimodal research has focused on text-image and text-video methods, a significant gap for educational applications.

The second study develops a cluster-then-classify methodology integrating categorical and continuous student data. Using records from 3,089 undergraduates, K-Prototypes clustering identifies five subtypes among non-returning students, and XGBoost and Gradient Boosting classifiers detect departure and assign subtype. Departure proves only weakly predictable from administrative and academic-performance data (non-returning F1 of 0.43), while subtypes are highly separable among students who do leave (macro F1 of 0.93 to 0.94); the binding constraint is detection rather than subtype assignment. Inverse-frequency class weighting outperforms synthetic oversampling, and a fairness audit on Pell eligibility finds that the most accurate model exhibits the smallest equalized-odds disparity.

The third study presents a privacy-preserving Remote Data Science framework: researchers from three universities of varying sizes develop classifiers on synthetic data, and the data owner executes them on one institution's private records with differential privacy applied to results. The framework achieves consistent performance (macro F1 of 0.690 to 0.695) under strict FERPA compliance, and its dual-server design suits smaller institutions with limited technical resources.

Together, these studies contribute methodological advances in educational data mining, a reproducible typology of undergraduate departure, and a practical path to inter-institutional analytics with responsible data practices.