Auto NAIVE

The Naive method provides automatic baseline forecasting by analyzing historical data and selecting the most appropriate basic forecasting approach. It serves as an essential benchmark for evaluating more sophisticated forecasting models.

How it works

The system automatically examines your data and selects one of four forecasting methods:

  1. Naive (Last Value):
    • When used: It is used for flat or stable data with no clear trend or seasonal pattern.
    • Forecast: It repeats the last observed value for all future periods.
  2. Drift (Trend):
    • When used: It is used when the data shows a clear upward or downward trend.
    • Forecast: It extends the historical trend into the future.
    • Example: If sales increased from 100 to 120 over 12 months, the forecast continues growing at +1.67 per month.
  3. SNaive (Seasonal Naive):
    • When used: It is used when a strong seasonal pattern is detected (weekly, monthly, quarterly, and so on).
    • Forecast: It repeats the last complete seasonal cycle.
    • Example: If December sales were 150, the next December forecast is also 150.
  4. SNaive + Drift (Seasonal with Trend):
    • When used: It is used when both a seasonal pattern and a growth or decline trend exist.
    • Forecast: It repeats the seasonal pattern while continuing the trend.
    • Example: It applies a seasonal pattern with 5% annual growth to each season.

Benefits

These benefits highlight the key advantages of using this method:

  • Automatic selection: No manual configuration is needed, as the system detects seasonality and trends automatically and selects the best approach for your data.
  • Transparent and explainable: Results are easy to understand and explain to stakeholders.
  • Performance benchmark: Provides a clear baseline to assess whether more sophisticated forecasting methods justify their complexity and cost.
  • Robust to noise: Resistant to overfitting and random fluctuations in historical data.

Usage tips

These guidelines outline the data requirements, data handling, and validation considerations for the method:

  • Minimum data: We recommend at least two seasonal cycles for reliable seasonal detection.
  • Data quality: The system automatically handles missing values and outliers.
  • Validation: Compare Naive forecasts with other methods. If performance is similar, consider simplifying your process.