Learning What Matters: Automated Feature Selection for Learned Cost Model in Parallel Stream Processing

Abstract

Learned cost models for parallel stream processing typically rely on hand-engineered features, which requires substantial domain expertise and manual effort. This paper presents an automated feature selection pipeline that identifies the features most relevant to predicting query performance in parallel stream processing, reducing the manual effort involved in building learned cost models without sacrificing prediction accuracy.

Publication
AIDB@VLDB 2025

Related