Predictive Analytics encompasses a variety of statistical techniques and Machine Learning solutions to build predictive models and visually interact with data in order to discover hidden insights and relationships, and thereby provide the basis for predictions about future events.
With the integration of SAP Hybris Marketing system and Predictive Analytics Library (PAL), we can use statistical methods to detect patterns and trends in historical data of banking and insurance, and use the information to predict customer behaviors.
In order to see how Hybris Marketing is using the power of PAL and how we can enhance the solution by leveraging the latest capabilities of PAL, we have developed a demo working on Hybris Marketing to predict buying propensity of customers.
Build the Front-end Demo
We created a predictive model in Hybris Marketing system to predict the buying propensity of customers. To define the context of the model, we selected a predefined predictive scenario Demo Banking Buying Propensity which includes the data source, the use case, and the applicable algorithms (implementation methods) for the predictive calculation.
After setting the status to In Preparation, we enhanced the model with model fits and scope. We created model fits with different implementation methods, different sets of predictors and cross validation. Then we trained the model and chose the best fit to publish the model. To restrict the validity of the model, we defined a set of countries and regions as scope for which the model was valid.
Before a predictive model can be created in front end, the predictive scenario has to be defined in IMG customizing at first. A predictive scenario defines the context of the predictive model, which includes the data source, the use case, and the applicable algorithms (implementation methods) and others.
The predictive scenario we used for the front-end demo above is Demo Buying Propensity Banking. It combines the predictive use cases such as prediction of buying propensity, the data source such as master and transactional data of CRM business partners, and the statistical methods to be used for the prediction calculation such as the logistic regression based on PAL. The data is provided by a set of standard SAP HANA information models and SQLscript procedures.
The predictive data source is addressed by the script-based calculation view sap.hana-app.cuan.cpred.demo.datafoundation/CA_CPRED_BANKING_DS_PAR. It provides the business data for the training set, which contains data that describes the actual customer behaviors. Then it’s used to detect patterns and trends for future prediction. The cross-validation and training of predictive models are based on a set of SQLscript procedures, which are used to call PAL functions to be used for predictive calculation.
PAL includes many new algorithms as well as enhancements to existing algorithms, such neural network. In order to enhance the predictive solutions of Hybris Marketing, we are going to create new calculation views to define the predictive data source, then create new SQLscript procedures for the validation and training of predictive models. At last, create predictive scenarios to leverage the latest capabilities of PAL.