Good Morning Bent
There is some detail here in the
PureCloud Resource Center, however to give you a little more information. To give you a little more context the model works in the following way.
There are four main components:
Data Management:
We focus heavily on historical data to predict change in values
- Historical Data Clean-up
- Missing Data Imputation
- Temporal Stream Aggregateion
- Distribution Transformation
Identify Outliers & Calendar Effect(s)
We evaluate properties of data and auto detect anomalies
- Identify 3 types of outliers: random pulse, temporal change, level shifts
- Identify holiday effect
- Identify lead / lag in calendar effect
Pattern Detection
Detect long-term patterns to ensure tactical accuracy
- Detects seasonality and trends
- Performs decomposition of time series data
- Ensure hierarchical temporal stream consistency via distributions
Best of the Best Model Selection
The process performs extensive cross-validation to select best method based on accuracy.
- Sophisticated fold strategy and scoring mechanism
- 27+ methodologies
The net result is using the 27 methodologies with thousands of determined configurations the forecast model runs through hundreds of thousands of forecasts and selects the most accurate one for the data presented.
We hope that clarifies how the capability works.
Regards
------------------------------
Cameron Smith
Sr Director, Product Management - Workforce Engagement Management
Genesys - Employees
cameron.smith@genesys.com------------------------------
Original Message:
Sent: 12-07-2018 06:39
From: Bent Slyngstad
Subject: Forecast: Automatic Best Method Selection
Anyone that can comment on what this method actually does ? What parameters are taken into account and to what degree ?
#WorkforceManagement
------------------------------
Bent Slyngstad
Atea AS
------------------------------