OCEG: an Interactive Visualization System for Classifier Ensemble Generation

• Mauro Diaz Universidad Católica San Pablo
• Guillermo Camara-Chavez UFOP
• Erick Gomez-Nieto Universidad Católica San Pablo

Resumo

Classification is a fundamental task in machine learning. However, individual models can struggle to deliver accurate results in various scenarios. Ensemble methods, which combine multiple classifiers, can enhance performance, but selecting and fine-tuning these models can be time-consuming. We introduce OCEG, a visual system designed to support the interactive exploration and composition of classifier ensembles. OCEG features a web-based interface that utilizes visual cues and pruning algorithms to highlight differences between models, automatically optimize ensemble parameters, and compare alternatives from a comprehensive perspective. We validate our system through quantitative experiments, a real-world case study, and user evaluations.

Palavras-chave: Visualization, Accuracy, Confusion matrices, Visual systems, Classification algorithms, Ensemble learning