Deciphering the Academic Web: Using Merged Bezier Curves to Visualize Educational Data Mining

TL;DR

Analyzing how students progress through university courses often results in a "hairball" of data. This paper presents a specialized multi-level visualization tool that replaces cluttered straight lines with Merged Bezier Curves. By grouping similar student paths and using thickness to represent volume, the authors achieved a 24% improvement in usability scores and a significant reduction in the time required for academic coordinators to identify critical course interdependencies.

The Problem: The "Textual Fog" of EDM

Educational Data Mining (EDM) is a goldmine for improving curriculum design. By applying sequential pattern mining to nine years of student records, researchers can identify patterns like: "Students who pass Calculus I and Physics I in Semester 1 are 80% more likely to pass Electromagnetism in Semester 2."

However, the output is often an unreadable list of thousands of textual strings. Existing tools like CourseVis or GISMO focus on individual student performance but fail to show the macro-level health of a curriculum. When you have over 1,000 patterns, a standard node-link diagram becomes an indecipherable mess of overlapping lines, known in the visualization community as "edge clutter."

Methodology: Layered Semantics and Merged Geometry

The authors suggest that the structure of the visualization must mirror the structure of the degree itself.

1. The Multi-Level Architecture

The system organizes data into vertical or horizontal layers representing semesters.

• Nodes (Courses): Represented by circles. Radius $\propto$ total students.
• Status Splitting: Circles are divided into green (success) and red (failure) semicircles, providing an immediate visual "health check" of a course.

2. The Innovation: Merged Bezier Connectors

The "Aha!" moment of the paper lies in how it handles interconnections. Instead of drawing a single line for every pattern (which leads to chaos), the authors implemented a merging mechanism:

• Bezier Curves: Unlike straight lines, curves allow for more organic "flow" and can be routed to avoid overlapping node labels.
• Visual Merging: If multiple patterns share a common origin and destination sequence, they are bundled together. The thickness of the curve is then adjusted to represent the aggregate number of students.

Figure 1: The multilayered visualization showing semesters as distinct tiers with course interrelations.

Experiments: Proving the "Curve" Advantage

The authors conducted a controlled study comparing "Simple Lines" (S1) against "Merged Bezier Curves" (S2) using scenarios of increasing complexity (from 8 patterns to 1,013 patterns).

Key Findings:

• Efficiency: For the most complex scenario (AL course with 1013 patterns), users were nearly 30% faster at identifying relevant patterns using Merged Curves.
• Accuracy: The accuracy of identifying the "top 3 most relevant patterns" remained near 100% for the curve method, whereas it dipped for the line method as clutter increased.
• Usability (SUS Score): The Merged Curves method scored 84.5, categorized as "Excellent," while the line method scored a mediocre 68.0.

Table: Comparison of time taken for Task 2. Note the massive gap in scenarios 4 and 5.

Critical Insight: The "Less is More" Principle

The success of this work isn't just about "pretty curves." It's about Information Abstraction. By performing what is essentially a "visual join" on the data at the rendering level, the system respects the human eye's limited bandwidth.

Limitations: The study uses a relatively small sample size (15 users), and while Bezier curves reduce clutter, they can still become "spaghetti" if the courses aren't strategically placed within the semester layers. Future work could benefit from automatic layout optimization or "force-directed" placement within layers.

Conclusion

This paper serves as a blueprint for anyone dealing with high-density sequential data. By moving from discrete lines to bundled flows, the authors have transformed a technical EDM output into a strategic tool for university deans. It proves that in data visualization, the way you connect the dots is just as important as the dots themselves.