classification-basics
Build and evaluate classification models — class balance, metrics (precision/recall/ROC), thresholds, and error costs. Use when predicting categories such as churn, fraud, conversion, or risk tiers.
- License 许可
- MIT
- 最近更新
- 2026-07-22
- Stars
- ★ 7
- Forks
- 1
01 前言 Front Matter
- name
- classification-basics
- description
- Build and evaluate classification models — class balance, metrics (precision/recall/ROC), thresholds, and error costs. Use when predicting categories such as churn, fraud, conversion, or risk tiers.
03 SKILL.md 正文
Classification Basics
Overview
Classification predicts discrete labels. Success depends on metric choice aligned to business costs, not only accuracy on imbalanced data.
When to Use
- Churn, conversion, fraud, and risk classification
- Lead scoring and propensity models
- Choosing thresholds for operational decisions
- Comparing classification algorithms baseline-to-baseline
Core Practices
- Inspect class balance and cost of error types
- Choose metrics that match the decision (precision, recall, F1, AUC, etc.)
- Use proper train/validation/test splits (and time-based splits when needed)
- Calibrate probabilities if decisions use scores
- Set thresholds from cost/benefit, not default 0.5
- Monitor drift after deployment
Principles
- 99% accuracy can be useless on 1% positive class
- False positives and false negatives have different prices
- Leakage from future data invalidates results
- Interpretability may be required in regulated settings
Verification
- [ ] Metrics match business error costs
- [ ] Evaluation method avoids leakage
- [ ] Threshold rationale is documented
04 同仓库更多 Skills
compliance-monitoring
Design compliance monitoring and testing programs that verify controls work in practice and surface issues before regulators do. Use when building monitoring plans, compliance testing, or continuous control monitoring.
complaints-handling
Operate complaints-handling processes that meet regulatory standards — logging, timely resolution, root-cause analysis, and fair outcomes. Use when designing or improving regulated complaints frameworks or analyzing complaint themes for conduct risk.
aml-kyc-basics
Apply AML/KYC fundamentals — customer due diligence, monitoring, suspicious activity escalation, and recordkeeping — for financial crime compliance. Use when designing onboarding controls, transaction monitoring concepts, or AML program components in financial services.
ai-model-compliance
Apply compliance controls to AI and automated decision systems — transparency, fairness, human oversight, documentation, and regulatory alignment. Use when deploying AI in regulated processes (credit, fraud, customer decisions) or responding to AI governance expectations.
data-quality-management
Manage data quality end-to-end — dimensions, rules, remediation, and prevention — beyond one-off cleaning scripts. Use when building sustained data quality programs or fixing systemic accuracy, completeness, or consistency problems.
data-lineage
Capture and use data lineage — where data comes from, how it transforms, and what depends on it — for impact analysis, debugging, and trust. Use when tracing metric origins, assessing change impact, or diagnosing broken pipelines.