2025 ASEE Annual Conference & Exposition

This paper is supposed to be the full paper under the category of Empirical Research.

This Empirical Research paper presents an analysis of a longitudinal dataset covering 10 years of student academic performance using statistical and machine learning methods, contextualized within the School of Electrical and Computer Engineering (ECE) at a large, public, research-intensive institution in the Southeast United States. This investigation expands upon the research presented at the 2024 ASEE Conference, which identified predictors of student academic success in an upper-level microelectronic circuits course from a smaller dataset with fewer predictors. In this study, we have expanded the analysis in several dimensions (i.e., time scale, predictor variables, and outcome variables) and have also developed machine learning models to predict student performance in core ECE courses. Altogether, this analysis indicates opportunities to help program leaders provide students with early, effective, and personalized support, enhancing academic success across the diverse backgrounds that students bring to their undergraduate studies. This work builds on prior work in engineering education modeling predictors of academic success through academic records (e.g., MIDFIELD), but is contextualized to a specific undergraduate program, which allows for a fine-grained interpretation of results that may be transferrable to other institutions.

The expanded dataset spans from Fall 2014 to Spring 2024 and includes data for approximately 1,600 students. The dataset includes students’ high school performance (e.g., whether taking advanced placement courses, SAT scores), academic records at the current institution (e.g., grades and number of attempts in various courses), and additional characteristics such as transfer credits, transfer history, gender, and ethnicity. Given the large scale of the institution and the diverse backgrounds of its students, the dataset exhibits significant heterogeneity across the aforementioned attributes. By applying data-mining techniques to this dataset, we can gain valuable insights into the factors that influence students’ academic performance throughout their time at the institution.

At the time of abstract submission, the analysis has concentrated on a 4-credit hour, junior-level course on semiconductor architecture that is notorious within the department for its difficulty. Of the 1,592 students who have taken the course in the study timeframe, 15.3% earned a final grade of a ‘D’, ‘F’, or ‘W’ and 19.0% earned ‘C’ in their first trial. Among these students, around 65% of students were direct matriculation, meaning that around a third of students were transfer or readmitted students, although this number has fluctuated over time. On average, transfer students demonstrated lower average performance compared to directly matriculated students. Students who took at least one advanced placement course in high school comprised 30.7% of all students, and we found a clear positive correlation between students’ average performance and the number of advanced placement courses they completed. Further data analysis of additional courses, along with regression and machine learning modeling, will be conducted and presented in the full paper.

We report these trends by student groups to present baseline conditions for this abstract, but we want to be intentional in avoiding any essentialist, deficit-based conclusions. Our goal is to engage in deeper analysis to identify influential factors behind these symptoms, facilitating the development of data-driven insights that will help program leadership make informed decisions about curricular developments and resource management.