About Prediction Accuracy
When you click (Details) to get
more information about the prediction data or the
historical data, you see the accuracy of the
prediction.
The amount of historical data available impacts the accuracy of the predictions; the more data the better. At a minimum, there should be at least twice the amount of historical data as the number of prediction periods. A ratio of three or more times the amount of historical data as the prediction periods is preferable. If not enough historical data is available at the time of prediction, a warning or error is displayed. Predictive Planning can detect seasonal patterns in the data and project them into the future (for example, spikes in sales numbers during holiday seasons). At least two complete cycles of data must be available to detect seasonality.
In addition, Predictive Planning detects missing values in the historical data, filling them in with interpolated values, and scans for outlier values, normalizing them to an acceptable range. If there are too many missing values or outliers in the data to perform reliable predictions, a warning or error message is displayed.
The prediction accuracy can also be affected by how noisy the data is. Even though a large amount of historical data may be available, the noise or random fluctuations in the data can obscure the underlying trend and cause the prediction accuracy to be low.
Using events for predictions improves the accuracy of the predictions and helps you plan ahead for the events and take advantage of opportunities by allowing you to see anticipated spikes and falls in the predicted data for specific events. Without events, spikes or falls in data are normalized and distributed over the prediction period, potentially leading to less accurate predictions.
In general, use these guidelines to determine a prediction's accuracy:
- 95 – 100%: Very Good. The historical data has a strong trend or seasonal pattern.
- 90 – 94.9%: Good. The historical data has a moderate trend or seasonal pattern.
- 80 – 89.9%: Fair. The historical data has a weak trend or seasonal pattern.
- 0 – 79.9%: Poor. The historical data has no detectable trend or pattern.