How the data predicted in “Sales Performance – Predictions”

Prediction Model for App Sales Performance – Predictions

App “Sales Performance – Predictions” uses predictive model “SALESVOLUME001” for prediction. (Help document )

This prediction model adopts regression analysis in PAI (Predictive Analytics Integrator). It relies on internal structured data (e.g, sales volume in billing documents and master data like sales org and product) from the last two years for training. External or unstructured data is not considered up to now, but might be incorporated as sources of training datasets in a future version.

The current model used the regression model in PAI.

Tips on Enhancing Prediction Accuracy

The quality of prediction results also depends on the following aspects of training datasets:

Data volume (at least 2000 data records)
Data continuity (continuous monthly data for more than one year)
Reasonable data distribution among your key business dimensions (according to your business cases)

For example, check if the business data is sufficient, continuous training data exists along the business dimensions. Ideally, we recommend that you conduct training after your system contains at least one-year continuous monthly data with more than 2000 data records. Otherwise, prediction results may not qualify for productive use.

Example business data that is not recommended for use:

In this example, there are only 5-month sales volume data and most of the sold-to parties do not have continuous data over the five months. Prediction results based on such training data has low credibility.

Prediction quality-related figures in app Predictive Models

In app “Predictive Models“(in business role template SAP_BR_ANALYTICS_SPECIALIST), you can train the model SALESVOLUME001. There are some APL(Automated Predictive Library) Indicators when you open a model version like screenshot below. For details about the indicators, see the “SAP HANA Automated Predictive Library Reference Guide”

Quality

The quality of your model version rated in star symbols (one – five stars, with five stars indicating the best quality). The Predictive Models app evaluates the quality of your model in terms of its prediction confidence and predictive power. Prediction confidence and predictive power give you a quality indicator from which you can decide whether to use your model version as is or do further training.

Predictive Power

Also known as KI. This is the quality indicator of the models generated using Automated Analytics. It corresponds to the proportion of information contained in the target variable that the explanatory variables are able to explain.

Prediction Confidence

Also known as KR. This the robustness indicator of the models generated using Automated Analytics. It indicates the capacity of the model to achieve the same performance when it is applied to a new datasets exhibiting the same characteristics as the training datasets

Training Record Count

An indication of the number of records the model version was trained on, after applying the training filter. Available when supported.