2024 ASEE Annual Conference & Exposition

Predicting Student Outcomes Through Artificial Intelligence for Intervention in Aerospace and Mechanical Students
Abstract
Academic intervention in underrepresented students during the early years of their engineering program plays a crucial role in improving program retention and academic success. These issues are particularly prominent in Science, Technology, Engineering, and Mathematics (STEM) fields, where many students opt to change majors due to difficulties in their programs. Artificial Intelligence (AI) has emerged as a powerful tool for predicting student outcomes and has the potential to revolutionize education. By leveraging AI, we aim to develop a framework that utilizes historical student data to predict future outcomes. The dataset used includes demographic information of students in an Aerospace and Mechanical Engineering (AME) program, as well as program completion and graduation status. The program is part of an urban, Carnegie R1 institution in the Southwestern United States. Our research question focuses on identifying the factors that contribute to accurate prediction of student outcomes in AME students. The two outcomes this work exhibits are whether or not the students graduated and if they completed the AME program or switched to another one. We employ the use of MATLAB's Statistics and Machine Learning Toolbox and Deep Learning Toolbox to train and test our neural networks. The Fitcnet command was selected to create a feed-forward neural network classifier. Initial results display a minimum level of accuracy of 92 percent for our trained models with variable predictor selection.The models not only successfully predict the program outcomes but also provide insight onto what predictor parameters are important and which ones do not affect the outcome. The initial work aims to identify key predictors for high-accuracy predictions as well as help identify and test more prediction factors used in the literature. Through future iterations of our model, we aim to further improve prediction accuracy by incorporating additional predictor data. Early identification of students at risk of changing or dropping from the program will enable targeted intervention and improve their chances of success. These initial iterations will serve as benchmarks, with the ultimate validation of our project relying on the performance of the AI model with data from current AME students upon their completion of the program.

Introduction
There continues to be a historical gap in STEM programs for both women and ethnic minorities, despite the progress made toward bridging this gap. Research has shown that underrepresented students who receive academic intervention within the first few years are more likely to persist in STEM programs [1, 2]. This highlights the importance of identifying students at a higher risk of dropping from these programs. In this paper, we aim to address this issue by leveraging the power of AI algorithms, specifically Neural Networks (NNs), to predict and identify patterns and trends related to student outcomes in STEM programs.

Previous studies utilizing machine learning and classification techniques have demonstrated their potential in predicting and understanding student performance in academic programs [3, 4, 5]. By leveraging these methodologies, along with key predictor groups such as Academic Achievement, Demographics, Environmental, and Psychological Variables, we seek to expand on this existing research [6, 7, 8, 9, 10, 11]. We will analyze institutional data divided into student demographics, academic preparation, outcomes, and additional data to be collected. Our ultimate goal is to develop a predictive model that can identify students in need of academic improvement, allowing educational institutions to monitor student performance and offer targeted support programs.

We believe that our research will contribute to the overall objective of promoting student success in STEM programs. By applying our findings at both macro and micro levels, we can foster a more inclusive and supportive learning environment. We will draw upon the work of Al-Doulat [12] to emphasize the need for collaborative efforts among all members of the institution and program leadership, as well as the research by Ramesh [13] that highlights the potential for intervention at an individual course level. This work focuses on the effect student demographics and personal data has on their performance and does not address the effect instruction quality and instructor training has on such performance. Efforts are being made to increase the amount of data available for each student.