Detecting Behaviors in
Student Progress Trajectories

Colm Howlin

Realizeit

@cp_h

colm.howlin@realizeitlearning.com

Charles Dziuban

University of Central Florida

@papuga

charles.dziuban@ucf.edu

Presented at Educational Data Mining 2019. July 2 - 5, 2019, Montréal, Canada

• Moved from a broadcast model to on-demand availability.

• Instructors want to know what their students are doing and what they should be doing.

• Are there student putting in lots of effort but getting nowhere? Are there students who aren’t really trying?

• Impact on student learning – attainment, retention of knowledge, grades.

Motivation

Use these controls to play and interact with the animation

Progress Trajectories

N=5044

51 online and blended courses

Nine terms - 2015 to 2018

Psychology, Spanish, College Algebra, various Computing courses, and Nursing.

Summarizing a Trajectory

Repeated Fuzzy Clustering

• Initial attempts using a single application of crisp or fuzzy clustering did not produce coherent groupings - Outlier behaviors were being missed

• He et al. [2016, 2017] used clustering to search for hidden communities in social networks.

  • First used clustering to discover the most apparent communities.

  • Then decreased the weights on the edges in the social network that represented hidden communities

  • Repeating the clustering uncovers previously hidden communities.

• Capture how well a cluster represented a student trajectory

  • Fuzzy clustering naturally lends itself to this

Repeated Fuzzy Clustering

Finding Outliers

• Students should be well described by at most two clusters – interpretable by instructors

• Radviz visualizes a clustering solution - points close to the center are evenly distributed among all clusters

Clustered Behaviors

3244 (64.3%)

915

(18.1%)

293

(5.8%)

340

(6.7%)

352 (5.0%)

Next Steps

• So what?

  • Stage 1: Instructor feedback → Stage x: Automated Interventions

  • Needs to be actionable

    • What is driving behaviours?

    • How consistent are behaviours?

    • What are the long term impacts?

• Algorithm

  • Include effort / engagement in the clustering

  • How robust is the algorithm to the choice of the number of clusters on each repetition?

  • The impact of the method to choose outliers

Colm Howlin

Realizeit

@cp_h

colm.howlin@realizeitlearning.com

Charles Dziuban

University of Central Florida

@papuga

charles.dziuban@ucf.edu

Detecting Behaviors in Student Progress Trajectories - Presented at Educational Data Mining 2019. July 2 - 5, 2019, Montréal, Canada

Contact