Computer-supported collaborative learning : predicting teamwork performance in collaborative project-based learning

Abstract

In today's connected world, the team formation of people to execute a project has been seen as a challenge in government agencies' public and private organizations. A small group of thoughtful and committed people performing in different roles could change the world for large enterprises. At the same time, it is hard to select an effective team whose members can work collaboratively. Pulse of the Profession, published by Project Management Institutes (2017), reported that failed projects always lacked (a) clearly defined objectives to measure progress and (b) poor communication between team members. Minimizing communication costs and maximizing trust levels are essential to improve the efficiency of team performance. This study's objectives required including how to formulate the problem and design the theoretical framework. The approach involved a five-step team formation model with related definitions, including initial team forming, depending on group size, agreement, role assignment, and team performance. In this project, we first analyzed students' academic records during the pre-processing stage to extract information about their English skills, leadership skills, communication skills, technology savvy skills, logical skills, and hardware skills. Nearly 851 records were collected, from students of three project-based subjects (each from an undergraduate programme), on their academic performance in the subjects relating to programming/technological study, hardware development, and generic IT study throughout three academic years. Then, these subjects were mapped to the relevant skills as the features, which are stored to form a data set and are used for training a machine learning model. In order to acquire a machine learning model as accurately as possible, based on the data set, we divided it into a training set and a test set to build and evaluate the model. Two-thirds of the data set were used as the training set, and the rest formed the test set. The test set was used to validate the model building, and data in the training set are excluded from the test set. The regression algorithm and Naïve Bayes were selected because they are commonly used machine learning algorithms and can produce promising performance. The assessment of member collaboration effectiveness was used the proposed communication cost algorithm and trusted direction algorithm. The proposed five-step team formation model is referenced from the group role assignment algorithm (GRA). Last but not least, the Predicting Teamwork Performance (PTPA) system was developed to help automatically identify each member's functional roles. Role assignment positively impacted team projects, while the role identification mechanism can assign team members responsibilities for some role(s) to enable learning. Self-assessment was used to identify team members' strengths and weaknesses so that team leaders could easily recognize suitable types of roles for each member. Three primary team performance indicators—" Good", "Pass", and "Marginal"—were reflected in the teamwork collaboration outcomes. The Predicting Teamwork Performance system reveals information about those outcomes through 1) individual performance indicator; 2) teamwork performance indicator; 3) personal skill sets results; 4) recommended skill sets improvements. The relationship between those indicators and functional roles was examined as analytical information for further project team formation.