Comparative Visualisation of High-Dimensional Data

Context

The thesis is Smit's most substantial technical project: a web-based system for visually comparing high-dimensional datasets using feature-wise binning and quantitative similarity measures.

Problem

High-dimensional datasets are difficult to compare because important differences can hide across many features and distribution shapes.

What I built

A web-based visual analytics tool with interactive visualisations, feature-wise distribution views, multiple binning methods and quantitative similarity comparisons.

Technical approach

• Implemented Gaussian, Voronoi and Generalised Adaptive Intelligent binning for feature-wise comparison.
• Integrated Earth Mover's Distance, SSIM and MS-SSIM as complementary similarity measures.
• Tested the system on synthetic and real-world datasets to explore how visual and metric comparisons behave together.

What I learned

The work connected interface design with analytical rigor: a useful comparison tool needs both readable visual structure and carefully chosen similarity measures.