Modelo preditivo de aprendizagem ativa em engenharia baseado na modelagem de equações estruturais com mínimos quadrados parciais (PLS-SEM)

Abstract

The Brazilian National Curriculum Guidelines (DCN) of the Undergraduate Engineering Courses of 2019 presented new pedagogical demands, one of them being the application of Active Learning (AL). Scientific studies on AL applied to Engineering courses has grown significantly in recent years and its results uncovered challenges and opportunities for future research. Classroom observation instruments designed for AL environments have emerged and have supported research to objectively assess the behaviors and attitudes that characterize such environments. However, there was a lack of predictability for Engineering Higher Education Institutions (EHEIs) regarding learning gains when applying AL techniques within a challenging process of changing teaching practices. In this context, the objective of this thesis was to propose a predictive mathematical model that demonstrates the relationship between the students' degree of learning and the application or not of AL techniques in the classroom (measured by the level of activity captured by an observation protocol). To achieve this objective, a strict and systematic methodological process was established, using controlled experimental research in an EHEI over three years, in two dimensions of analysis. The first one, intraclass, used a repeated measures experimental design to demonstrate the probable causeeffect relationship in a two-level one-factor approach. In addition, it allowed a qualitative analysis of AA application in individual courses. The second, interclass, involved independent class samples in subsequent semesters and used Partial Least Squares Structural Equation Modeling to test and identify the best predictive model for learning based on the application of AL. The intraclass results demonstrated, in a positive cause-effect relationship, that the global average academic performance was 14% better in the post-AL assessment, compared to the first, without the application of AL techniques, representing 40% of the standard deviation of the grades. In addition, the individual analysis of performance in each of the courses revealed the most and least successful strategies and allowed to recommend those most viable AL strategies for specific groups of courses. In the interclass dimension, the improvement was 10% and the PLS-SEM predictive model was positively validated by several performance indexes, demonstrating a significant and non-linear positive relationship between the latent constructs, with a moderate to high relevance for learning prediction (𝑄2 > 0.344). In the demonstration of the predictive relevance, the best-fitting curve of the relationship allowed, from an average score between 0 and 35.97 in the level of AL adherence (NAA) to predict an average Learning score (AP) between 45.89 and 74.90 on the scale of performance degrees. β coefficients were positive and significant, with p values < 0.01. The systematic methodological design and the results obtained are intended to be the main contributions of this research to the literature and to the latent discussion of the effectiveness of active learning methods in Engineering Education.