Customer, Sales & Demand Analytics with SAP Analytics Cloud (SAC)
- Time-Series Forecasting
Whilst building time-series model for forecasting on SAP Analytics Cloud (SAC) platform, entity are defined so that the model type exposes dimension (attributes) as single entities and enables configuring target variable for prediction including date dimension.
In my context, the time-series model used for prediction or forecasting on the target variable that i build comprises of these dimension(s) and measures.
- Exploratory (ie: Independent variables) —
- Order Date (Date type)
- Entity (Product Category, Product sub-category)
2. Target (ie: Dependent Variable)—
- Order Quantity (ie: Order Demand)
- Business Understanding
Supply Chain planning based on the order quantity (demand) to ensure inventory replenishment are adequate for manufacturing process to ensure timely order fulfillment. Due to raw material availability subjected to order lead-time, inventory replenishment is challenging as any manufacturing firm encountering inventory stock-out issues due to delinquent delays in supplier’s deliveries is very costly impact on production process. As such, demand forecasting or prediction requires skillful computation to address buffer stock (safety stock) to mitigate stock-out issue due to supplier’s late deliveries or firm’s internal production over-consumption issues.
2. Data Encoding & Preparation
Ensure proper transformation or encoding variables with data type & statistical type (ie: continuous, ordinal, nominal, textual) are applied to the relevant entity, target and date variables.
- Exploratory (ie: Independent variables) —
- Order Date → Data Type: Date; Statistical Type: Continuous
- Entity (Product Category, Product sub-category) → Data Type: String; Statistical Type: Nominal
2. Target (ie: Dependent Variable) —
- Order Quantity is associated with Aggregation Type: SUM;
→ Data Type: Integer; Statistical Type: Continuous
3. Data Understanding
To ensure data quality, SAP SAC has incorporated “validation” button features which highlights or alerts any data errors detection for all variables in the dimensions & measures. Hence, “validation” checks has been performed on full dataset for errors detection before i proceed with data pre-selection or filtering for segmentation modelling.
4. Modelling Phase
Challenges encountered during the model training phase is achieving success for the entity variable of various values. As shown in the image below, training issues by Entity.
As such, dataset filtering on the entities shall be performed to achieve segmentation modelling to overcome this training issue.
Based on the entities that i have selected (ie: product category, product sub-category) which has been trained successfully in this time-series regression model, the model generates predicted(forecast) values of target variable & residual error for all the number of forecast period(s) that i input.
Residual Error Min. or Max may be displayed depending on other parameters that varies accordingly.
As shown below, the time-series regression model displays both the forecast & actual values of target variable (ie: order quantity) for the 6 forecast period. Error Max. or Min. values also depending on the deviation or variance from the actual values.
In my opinion, SAC (SAP Analytics Cloud) requires lesser mouse-click to display both data visualization & statistical data (ie: target values, residual error etc) compare to SAP Predictive Analytics application as shown below.
The time-series forecasting chart below uses dataset which are available from the training server, hence below chart is just for illustration purpose to compare the graphical-user interface difference between SAC (SAP Analytics Cloud) & Predictive Analytics.
● The red line is the trend.
● The blue line is the model forecast.
● On the right-side we will see the 6 weeks forecast and the zone of possible error where the predicted signal could be. The error bars are only displayed for the forecasts.
SAP Predictive Analytics is business intelligence software from SAP that is designed to enable organizations to analyze large data sets and predict future outcomes and behaviors. For example, SAP Predictive Analytics can help make sense of big data and the Internet of Things by building predictive analytics models to identify unforeseen opportunities, better understand customers, and uncover hidden risks.
SAP Predictive Analytics is a statistical analysis and data mining solution that enables building of predictive models to discover hidden insights and relationships in our data, from which predictions about future values can be made or recommender can identify the highest number of links between nodes used in link association network analysis.
SAP Predictive Analytics combines SAP InfiniteInsight and SAP Predictive Analysis in a single desktop installation. SAP Predictive Analytics includes two user interfaces, Automated Analytics and Expert Analytics.
Other related articles written by Kuosheng:
