1. Abstract
Systematic and business-aware treatment of missing values preserves data continuity while avoiding artificial bias. Proper gap handling ensures that correlations and forecasts remain meaningful and reliable.
2. Context
Apply this best practice before uploading data into Foresight or using it in Discover, Workbenches, or Predict. Foresight requires complete time-series data for optimal modeling performance.
3. Content
3.1 Why It Matters
Missing values distort correlations, truncate model training windows, and can force Predict to rely on shorter histories than intended. Improper gap filling can be just as harmful as leaving gaps unaddressed, especially when business context is ignored.
3.2 How to Apply
Use a structured approach to classify and treat missing values:
- Identify the type of missingness
- Random: Isolated missing points with no clear pattern
- Systematic: Repeating gaps (e.g., delayed reporting)
- Business-related: Gaps caused by known events (shutdowns, outages)
- Apply appropriate methods
- Short gaps: Linear interpolation
- Seasonal data: Seasonal-naïve (same period prior year)
- Longer gaps: Related internal metrics or external benchmarks
- Validate results visually and logically
- Confirm smooth transitions around filled values
- Ensure imputed points align with known business behavior
- Document all assumptions
- Record why a method was chosen
- Flag imputed periods for future reference
3.3 Example
A production dataset is missing July 2020 due to a system outage. The analyst interpolates between June and August values, validates the result against production capacity records, and documents the adjustment as “System Outage – July 2020.”
3.4 Common Pitfalls
- Forward-filling long gaps without justification
- Applying the same method to all missing values regardless of cause
- Failing to visually inspect the results
- Not documenting which values were imputed
3.5 Expected Results
- Continuous, complete time series ready for Discover and Predict
- Reduced risk of biased coefficients or truncated training windows
- Clear documentation that supports transparency and reuse
