CostCla a cost-sensitive classification library
summary
CostCla is a Python library (scikit‑learn pandas numpy) that implements example-dependent cost matrices and cost-aware learners and claims big profit gains over standard classifiers. On a credit dataset, applying a base minimum risk classifier raises measured savings from about 17% to ~42%, and training cost-aware models (including a random patches ensemble) pushes savings to roughly 50%. The code is available, integrates with usual ML pipelines, and demonstrably increases business profit versus cost‑insensitive baselines.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 2:36Credit scoring importance
- 5:48Demo with Kaggle dataset
- 8:38Traditional metrics limits
- 11:30Cost-sensitive classification
- 14:18Building cost matrix
- 16:12Calculating savings
- 18:41Example-dependent costs
- 20:46Minimum risk classifier
- 22:54Training for financial cost
- 24:45Cost-driven model development
- 25:26Introducing CostCla
- 26:56Future work
- 27:19Q&A
related talks
48mOptimize the Right Thing
35mUncertainty Unleashed
28mGet the best from your scikit-learn classifier
44mHandling predictive uncertainty in Machine Learning
26mCausal inference with Causallib
34mMaximizing Credit Card Profitability with ML–Driven CLV
36mThe Social Cost of Strategic Classification
42mCreating correct and capable classifiers