When Traditional QA Fails for Machine Learning
from Software Testing with Fexingo: QA, Automation, and Reliable Software Engineering Hosted by Fexingo
In this episode, Lucas and Luna dive into why conventional software testing approaches break down for machine learning models. They explore a 2024 Google study revealing that 70% of ML production failures stem from data issues, not code bugs. The conversation covers the differences between deterministic and probabilistic testing, the critical role of data validation and schema enforcement, and how teams can detect model drift and data skew in production. Using a real-world credit scoring model example, they explain why relying solely on accuracy metrics can be dangerous and how continuous monitoring with tools like Great Expectations and Alibi Detect can catch failures early. Perfect for QA engineers and data scientists looking to build reliable ML pipelines. #MachineLearningTesting #MLQA #DataValidation #ModelDrift #SoftwareTesting #QualityAssurance #AI #DataScience #GreatExpectations #AlibiDetect #ProductionMonitoring #ContinuousTesting #Technology #FexingoBusiness #BusinessPodcast #ReliableML #DataQuality #MLOps Keep every episode free: buymeacoffee.com/fexingo
Embed this episode
No transcript yet.