FELA: Multi-Agent LLM System for Automated Feature Engineering on Event Logs
A team of researchers has introduced FELA (Feature Engineering LLM Agents), an evolutionary multi-agent system that utilizes large language models to independently identify significant features from intricate industrial event log data. These event logs, which capture detailed user interactions and system occurrences, present challenges due to their large volume, high dimensionality, varied data types, and complex temporal or relational frameworks. Current automated feature engineering techniques, such as AutoML and genetic algorithms, struggle with limited explainability, inflexible processes, and inadequate adaptability. FELA combines LLM reasoning and coding skills with a self-evolution approach guided by insights to address these challenges. It is adept at managing the complexity and diversity of industrial event logs, generating effective features autonomously. The research can be found on arXiv with the identifier 2510.25223.
Key facts
- FELA stands for Feature Engineering LLM Agents.
- It is a multi-agent evolutionary system for feature engineering.
- The system uses large language models (LLMs) for reasoning and coding.
- It addresses challenges of industrial event log data: large scale, high dimensionality, diverse data types, complex temporal/relational structures.
- Existing approaches like AutoML and genetic methods have limitations in explainability, flexibility, and adaptability.
- FELA employs an insight-guided self-evolution paradigm.
- The paper is published on arXiv with ID 2510.25223.
- Event logs record fine-grained user actions and system events.
Entities
Institutions
- arXiv