DataTerminology.comNEW
Also known as: KPI, Success Indicator
A performance metric is a quantitative measure used to evaluate how well a system, model, or process meets objectives.
DESCRIPTION
Performance metrics quantify outcomes to assess progress toward objectives in analytics, operations, and machine learning. In business, KPIs track strategic goals; in ML, metrics compare predictions to ground truth. Good metrics are relevant, reliable, and resistant to gaming. Selection depends on context—e.g., F1 for imbalanced classification or RMSE for regression. Clear definitions, baselines, and thresholds enable monitoring, alerting, and continuous improvement.
ALTERNATIVE DEFINITIONS
Wikipedia
A performance indicator or key performance indicator (KPI) is a type of performance measurement.
Defines KPIs as metrics used for performance measurement.
IBM
Key performance indicators (KPIs) are quantifiable measures of performance over time for a specific objective.
Highlights quantification and linkage to specific objectives over time.
RELATED TERMS
Key Performance Indicator Objective Threshold Baseline Model Model Evaluation Accuracy Precision Recall RMSE
COMPETING TERMS
Proxy Metric Vanity Metric
RELATIONS
- Broader: Metrics
KNOWLEDGE AREAS
Data Science AI Data Warehousing & BI Testing & Validation
TARGET AUDIENCE
Data Scientist BI Specialist Data Analyst AI Engineer Chief Data Officer AI Product Manager
EXTERNAL SOURCES
- Wikipedia - Performance indicator
- scikit-learn - ML model evaluation metrics guide
- Microsoft Learn - Azure ML metrics concepts
TOP QUESTIONS
Why are performance metrics essential in analytics and ML?
They translate objectives into measurable signals, enabling comparison, optimization, and accountability. Without clear metrics, teams risk optimizing the wrong outcomes.
Category: Relevancy
How do you choose a performance metric for classification?
Consider class imbalance and business costs. Accuracy may mislead; F1, ROC AUC, or PR AUC often better reflect performance.
Category: Best practices
What is a vanity metric versus a performance metric?
Vanity metrics look impressive but don’t drive decisions. True performance metrics align with objectives and behaviors you want to encourage.
Category: Misunderstanding
How do baselines and targets relate to performance metrics?
Baselines establish current performance; targets set desired future levels. Together, they support planning and evaluation.
Category: Management
Can a single metric capture overall performance?
Rarely. A balanced set reduces blind spots and metric gaming. Use composite dashboards with context.
Category: Opinion
How do thresholds and alerts use performance metrics?
Define acceptable ranges and trigger alerts when metrics deviate, enabling rapid response to issues.
Category: Operations
What is the risk of proxy metrics?
They may diverge from true objectives, incentivizing undesired behavior. Regularly validate proxies against outcomes.
Category: Risks
How often should metrics be reviewed?
At least each planning cycle; more often for operational metrics. Metrics evolve as objectives and environments change.
Category: Management
What’s a common mistake with performance metrics?
Changing metrics post hoc to justify results. Define metrics upfront and keep governance around changes.
Category: Mistakes