Stable 27 - Your Training Data is Bad and You Should Feel Bad

Explore the critical issue of poor training data in machine learning through this 22-minute conference talk from Derbycon 2018. Delve into the motivation behind the research, methodologies employed, and key findings regarding classifier performance. Examine tweet-only and user-only features, and understand their impact on model accuracy. Gain insights into future research directions and participate in a thought-provoking discussion on improving data quality for more reliable machine learning outcomes.

Intro
MOTIVATION
QUESTION
METHODS (CONT.)
RESULTS: CLASSIFIER PERFORMANCE
FINDING #2
FINDING #3
BOTTOM LINE
FUTURE RESEARCH
DISCUSSION
FEATURES: TWEET-ONLY
FEATURES: USER-ONLY