S10-O/LT8-5 - Genetic Algorithm optimization of teams for heterogeneity3. Research Work In Progress
1 University of Cincinnati
This work in progress study aims at developing a system to designate teams to relatively large groups of students in a classroom setting. It is motivated by recent changes in the ABET Criterion 3 accreditation guideline that requires students to demonstrate "an ability to function effectively on a team whose members together provide leadership, create a collaborative and inclusive environment, establish goals, plan tasks, and meet objectives." Outside of accreditation guidelines, team-based learning is becoming the prescribed pedagogical tool to enhance student collaboration and prepare them for professional work environments upon graduation. One important factor in effective team-based learning is to ensure teams are balanced in their abilities and characteristics across team members. Demographical memberships of students can also play an important role in team dynamics and impact the overall success in content learning. Instructors usually assign teams to students by manually looking into their abilities from previous grades and other performance factors. This can then become tedious when aiming to maintain uniform heterogeneity in classrooms of students coming from diverse backgrounds of knowledge, skills, ethnicity, and gender.
While some studies have presented the use of computer-aided tools, visualization categorization, and other Artificial Intelligence techniques, this work proposes the use of the Genetic Algorithm (GA) to form teams optimized for heterogeneity. The Genetic Algorithm presented uses a discrete integer-based chromosome representation and groups alleles to represent each team. Standard GA operators enable the algorithm to be adapted to any optimizing criteria deemed fit by the instructor. The algorithm presented in this work uses self-reported competency data on 3 different computational skills for 1300 students enrolled in a first-year engineering design thinking course divided into over 20-course sections of strength 40-60 each. Net team scores for each computational tool are calculated and the variation across each skill minimized by the fitness function for individual sections. Constraints based on gender and ethnicity are applied to minimize demographical imbalance between teams. The algorithm is tested for all sections in the Fall 2019 cohort to give consistent team configuration. Discussions on the stability and validity of configurations generated are discussed. Future work include methods that use a fuzzy logic decision-making tool for multiple criteria optimization.