Driftage

Driftage implements a multi-agent system framework to detect and interpret concept drift in time-series data, enabling robust drift management for applications such as electromyography muscle activity monitoring.

Key Features:

• Multi-agent architecture: Implements a multi-agent system framework that structures concept drift detection into interacting agents.
• Agent specialization: Divides the drift detection task into manageable segments handled by specialized agents.
• Distributed detection responsibilities: Distributes responsibilities of concept drift detectors across agents to create modular detection components.
• Interpretability and explainability: Enhances interpretability and explainability of individual detection components and their decisions.
• Integration with knowledge bases: Enables interaction between concept drift detectors and external knowledge bases to enrich analysis.
• Empirical reduction of false positives: Demonstrated reduced false-positive detections in a muscle activity monitoring case study using electromyography.

Scientific Applications:

• Electromyography muscle activity monitoring: Applied to detect and interpret concept drift in electromyography-based muscle activity monitoring, lowering false-positive detections.
• Concept drift management in non-stationary data: Applicable to maintaining machine learning model validity under evolving data distributions across bioinformatics and other domains.

Methodology:

The framework partitions concept drift detection into specialized agents that distribute detection responsibilities and enables interaction between concept drift detectors and external knowledge bases.