Machine Learning Techniques for Class Imbalances & Adversaries
summary
A practical survey that reports what is actually tried and what works: the Data Innovation Lab tests SMOTE sampling and finds synthetic minority interpolation helpful, applies time-decay weighting to downweight stale fraud signals, uses PCA and neural autoencoder outlier scores for anomaly features, ensembles scores into a meta‑learner (including non‑negative logistic coefficients), experiments with genetic algorithms, and runs GLRM via H2O for high‑cardinality fields. Expect concrete tool names, implementation notes, failure modes (e.g. SMOTE in high dimensions), and where methods improved production models.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 2:28Three key techniques
- 3:30Sampling methods
- 7:18Feature engineering
- 13:15Modeling strategies
- 16:23Q&A
related talks
39mBeyond Bias: Crafting Fairness in Pythonic Machine Learning Systems
47mEthical Machine Learning
43mToward Human-Centered Machine Learning
44mResampling techniques and other strategies
30mMachine Learning: Beyond the Tutorial
46mA look at the methods to detect and try to remove bias in ML models
50mEthical Machine Learning
36mMachine Learning Fairness: Lessons Learned