Identifying Factors that Enable Pinpointing At-Risk Students in a Programming Course

Haroon Malik; David A Dampier

Download Paper | Permalink

Conference: 2022 ASEE - North Central Section Conference
Location: Pittsburgh, Pennsylvania
Publication Date: March 18, 2022
Start Date: March 18, 2022
End Date: April 4, 2022
Tagged Topic: Diversity
Page Count: 4
DOI: 10.18260/1-2--39248
Permanent URL: https://peer.asee.org/39248
Download Count: 316

Paper Authors

biography

Haroon Malik Marshall University

visit author page

Dr. Malik is an Associate Professor at the Department of Computer Sciences and Electrical Engineering, Marshall University, WV, USA.

visit author page

biography

David A Dampier Marshall University

visit author page

Dr. Dave Dampier is Interim Dean of the College of Engineering and Computer Sciences and Professor in the Department of Computer Sciences and Electrical Engineering at Marshall University. In that position, he serves as the university lead for engineering and computer sciences. Prior to joining Marshall, Dr. Dampier served as Professor and Chair of the Department of Information Systems and Cyber Security at U.T. San Antonio, and Director of the Distributed Analytics and Security Institute at Mississippi State University. Prior to joining MSU, Dr. Dampier spent 20 years active duty as an Army Automation Officer. He has a B.S. Degree in Mathematics from the University of Texas at El Paso, and M.S. and Ph.D. degrees in Computer Science from the Naval Postgraduate School. His research interests are in Cyber Security, Digital Forensics and Software Engineering.

visit author page

Download Paper | Permalink

Abstract

Motivation ⸺ In the USA, students approach the study of computing in higher education in increasing numbers from a wide variety of backgrounds and disciplines. Given its importance, it is disappointing to realize that the teaching of programming (perhaps, more accurately, the learning of programming) is a perennial problem. Experienced teachers are all too familiar with the struggles of new students as they attempt to come to terms with this most fundamental area of expertise. Many teachers will have seen students choose course options (or even change degree programs) to avoid programming. Much of the existing research in the computing education literature at least focus on new and exciting ways to teach programming and model student performance to customize the learning environment(s); especially in online programming courses. The objective of the paper/study is to leverage machine learning to understand the factors that make learning to program so very difficult for many students. Moreover, leverage the sensitivity analysis to rank the factors based on their criticality towards augmenting the student performance to program, thus providing actionable insight to instructors to aid students to be successful in a programming course. Methodology ⸺ Our work identifies factors that influence a student's aptitude to program. Thus, we model our work as a classification problem. We chose to use Advance Decision Trees (ADT) classification technique, among others, i.e., Support Vector Machines, Neural Networks and Regression due to the fact that decision trees produce explainable models which are essential in (a) understanding the programming failure phenomena and (b) to find out the essential attributes in determining the likelihood of student failing the programming course. The study spanned over 220 undergraduate students from different disciplines ranging from computers science, electrical and computer engineering, math, statistics, music, nursing and humanities, over two years enrolled in their first programming course. We used 10-fold cross-validation to train and test the machine learning classifiers on the fifty attributed. The attributes are harvested from the course management system, students survey and peer-reviewed evaluations spanning over six dimensions; including Personal⸺ motivation, mode, study time, sleep, habits (drink, smoke, play), etc.; Family ⸺ income level, marital status, number of siblings, support, first-gen, etc.; Peer-related ⸺ classmates, study group members, friends, etc.; Subject/contents ⸺ course structure, grading policy, textbook, etc.; Institutional agent ⸺ faculty, advisor, staff, etc.; Social-culture ⸺class participation, friends, , group player, etc. Results ⸺ The work-in-progress (WIP) case study shows that we can predict with an accuracy of 89 %; the likelihood of a student failing the programming course. We performed a sensitivity analysis to determine the most critical attributes that make programming challenging to learn. Our analysis shows that “Motivation”, “Class Attendance”, and ‘Study Time’ are the most important attributes. Whereas, the family income level and being a first-generation student is the least important. Threats to the Validity ⸺ Our work identifies the factors that likely botch student performance in a programming course. The work in no way pinpoints the rationale of the poor performance in a programming course. Moreover, the work is constrained by the attributes that are harvested from the course management system and from course evaluations, i.e., preset questionnaire ⸺what the department/college choose to collect as part of course evaluations and not what student feel is important.

Citation
Format

Malik, H., & Dampier, D. A. (2022, March), Identifying Factors that Enable Pinpointing At-Risk Students in a Programming Course Paper presented at 2022 ASEE - North Central Section Conference, Pittsburgh, Pennsylvania. 10.18260/1-2--39248

TY  - CPAPER
AB  - Motivation ⸺ In the USA, students approach the study of computing in higher education in increasing numbers from a wide variety of backgrounds and disciplines. Given its importance, it is disappointing to realize that the teaching of programming (perhaps, more accurately, the learning of programming) is a perennial problem. Experienced teachers are all too familiar with the struggles of new students as they attempt to come to terms with this most fundamental area of expertise. Many teachers will have seen students choose course options (or even change degree programs) to avoid programming. Much of the existing research in the computing education literature at least focus on new and exciting ways to teach programming and model student performance to customize the learning environment(s); especially in online programming courses. The objective of the paper/study is to leverage machine learning to understand the factors that make learning to program so very difficult for many students. Moreover, leverage the sensitivity analysis to rank the factors based on their criticality towards augmenting the student performance to program, thus providing actionable insight to instructors to aid students to be successful in a programming course. 
Methodology ⸺ Our work identifies factors that influence a student&#39;s aptitude to program. Thus, we model our work as a classification problem. We chose to use Advance Decision Trees (ADT) classification technique, among others, i.e., Support Vector Machines, Neural Networks and Regression due to the fact that decision trees produce explainable models which are essential in (a) understanding the programming failure phenomena and (b) to find out the essential attributes in determining the likelihood of student failing the programming course. The study spanned over 220 undergraduate students from different disciplines ranging from computers science, electrical and computer engineering, math, statistics, music, nursing and humanities, over two years enrolled in their first programming course. We used 10-fold cross-validation to train and test the machine learning classifiers on the fifty attributed. The attributes are harvested from the course management system, students survey and peer-reviewed evaluations spanning over six dimensions;  including Personal⸺ motivation, mode, study time, sleep, habits (drink, smoke, play), etc.; Family ⸺ income level, marital status, number of siblings, support, first-gen, etc.; Peer-related ⸺ classmates, study group members, friends, etc.; Subject/contents ⸺ course structure, grading policy, textbook, etc.; Institutional agent ⸺ faculty, advisor, staff, etc.;  Social-culture ⸺class participation, friends, , group player, etc.
 Results ⸺ The work-in-progress (WIP) case study shows that we can predict with an accuracy of 89 %; the likelihood of a student failing the programming course. We performed a sensitivity analysis to determine the most critical attributes that make programming challenging to learn. Our analysis shows that “Motivation”, “Class Attendance”, and ‘Study Time’ are the most important attributes. Whereas, the family income level and being a first-generation student is the least important. 
Threats to the Validity ⸺ Our work identifies the factors that likely botch student performance in a programming course. The work in no way pinpoints the rationale of the poor performance in a programming course. Moreover, the work is constrained by the attributes that are harvested from the course management system and from course evaluations, i.e., preset questionnaire ⸺what the department/college choose to collect as part of course evaluations and not what student feel is important.


AU  - Haroon Malik
AU  - David A Dampier
CY  - Pittsburgh, Pennsylvania
DA  - 2022/03/18
PB  - ASEE Conferences
TI  - Identifying Factors that Enable Pinpointing At-Risk Students in a Programming Course
UR  - https://peer.asee.org/39248
DO  - 10.18260/1-2--39248
ER  -