Machine Learning and Student Performance in Teams

Ahuja, Rohan; Khan, Daniyal; Tahir, Sara; Wang, Magdalene; Symonette, Danilo; Pan, Shimei; Stacey, Simon; Engel, Don

This project applies a variety of machine learning algorithms to the interactions of first year college students using the GroupMe messaging platform to collaborate online on a team project. The project assesses the efficacy of these techniques in predicting existing measures of team member performance, generated by self- and peer assessment through the Comprehensive Assessment of Team Member Effectiveness (CATME) tool. We employed a wide range of machine learning classifiers (SVM, KNN, Random Forests, Logistic Regression, Bernoulli Naive Bayes) and a range of features (generated by a socio-linguistic text analysis program, Doc2Vec, and TF-IDF) to predict individual team member performance. Our results suggest machine learning models hold out the possibility of providing accurate, real-time information about team and team member behaviors that instructors can use to support students engaged in team-based work, though challenges remain.