ML Classifier Visualisation Tool

Demystifying Machine Learning

Built as part of a second-year group project at the University of Nottingham, this desktop application is designed to visualise how ML classifiers interpret data — making the black box of machine learning more transparent for students, educators, and researchers alike.

• ✔ Loads ONNX models with no extra config
• ✔ Renders decision boundaries in 2D and 3D
• ✔ Trains & evaluates models directly in-app
• ✔ Uses gRPC and Protocol Buffers to bridge Python backend and Java/OpenGL frontend
• ✔ Entirely local — no data leaves your device

Input images are converted into a training format that allows classifiers to interpret complex shapes — even faces — as mathematical boundaries using logical expressions.

• • Supports arbitrary expressions
• • Scales to 100,000+ points in real time
• • Fully interactive input and zoomable canvas

Our goal was to create clarity around classifier decision-making — enabling users to zoom into areas of interest, adjust sampling resolution, and intuitively see where models perform well or fail.

🧠 Technologies & Engineering

• • Python (model training, logic, data sampling)
• • Java with OpenGL (interactive UI rendering)
• • gRPC and Protocol Buffers for inter-language comms
• • ONNX for ML model portability

This architecture gave us real-time performance across systems while separating concerns cleanly — ideal for future scalability.

Demo Video

We’ll be uploading a full demo video walkthrough shortly. This will showcase our classifier running on different datasets and visually updating in real time as data points are adjusted.

Reflection

This tool helped us win top marks in our Software Engineering module (achieving a First), and more importantly, taught us the intricacies of software collaboration, UI design, and real-world ML system deployment.

Huge thanks to my team and to Dr. Andrew Parkes for mentorship and guidance throughout.