Domain-Specific Outlier Rules
What you'll learn: How to apply business logic and real-world context to create custom outlier rules that statistical methods alone might miss.
Why Context Matters More Than Statistics
You've learned several statistical methods for detecting outliers—Z-scores, IQR, and MAD. But here's the truth: a data point can be statistically normal and still be an outlier, or vice versa, depending on what the data represents.
Domain-specific outlier rules use your understanding of the business, industry, or field to define what values are actually impossible, illogical, or require investigation—regardless of what the math says.
Real-World Examples
Age in customer data: Even if someone's age of 150 falls within 3 standard deviations (passing the Z-score test), you know it's invalid—humans don't live that long.
Temperature readings: If you're monitoring a freezer that must stay below 0°C, a reading of 5°C is an outlier by definition, even if it's close to the mean due to equipment malfunction.
Transaction amounts: A $0.01 purchase might be statistically normal, but if you sell industrial equipment where the minimum order is $10,000, it signals a data entry error or system glitch.
Time-based rules: Sales of winter coats in July might look like outliers, but they could be legitimate clearance sales or southern hemisphere orders.
Building Your Rules
Combine what you know about:
- Physical constraints (negative ages, impossible dates)
- Business rules (minimum/maximum prices, valid categories)
- Domain knowledge (seasonal patterns, regulatory limits)
- Expert input (what practitioners say "never happens")
Key Takeaway: Statistical outlier detection is your starting point, but domain expertise defines which outliers matter and which rules override the numbers—real data science always requires understanding the context behind the data.