Explaining Detection Methods

Objective

After completing this lesson, you will be able to explain detection methods.

Introduction

Facts about detection methods:

Detection methods are used to search for irregularities in your data.
Within detection methods, you can define a business logic that is used to determine if an incident, such as a claim, tax declaration, or a bank transfer, is a potential fraud. A detection method can be used in multiple detection strategies and can be used for online detection as well as for mass detection.
A detection method represents a single executable step that detects fraud candidates in a given set of detection objects and assigns scores to them.
Several detection methods can be sequentially applied on the same data set by grouping them with a detection strategy. The result of a detection method is independent from the result of any other detection method used within the strategy.

Since SAP BIS supports different types of detection methods, you can use the following algorithm to decide what kind of automated controls you need to use as a starting point.

A flowchart laying out the decisions needed to determine if machine learning is the solution.

The following types of detection methods can be created:

Detection methods based on ABAP-managed database procedures
You can use ABAP-managed database procedures, which are created in the back-end system in ABAP and contain the SQLScript-logic that is executed in the SAP HANA database.
Detection methods used for address screening
You can use detection methods to screen names and addresses in business data against provider-defined lists.
Detection methods used for predictive detection
You can use your historical data to create detection methods that are based on a predictive model.
Detection methods based on SQLScript procedures in the SAP HANA Repository
You can use SQLScript procedures which are defined in the SAP HANA Repository and contain the logic that is also executed in the SAP HANA database.
Detection methods based on business rules

You can use detection methods which use either a text rule or decision table using HANA Rules Framework.

Note

Please note, that usage of detection methods based on the HANA Repository and HANA Rules Framework is not recommended. They remain in the solution for backward compatibility purposes only and are not to be used for any new implementations.

For this reason, they are not covered in this course.

Detection methods based on ABAP-managed database procedures

These detection methods are used to identify suspicious transactions and master data by programming the detection logic explicitly in the system.

Optionally, parameter can be used to provide the end-user with additionally flexibility when defining detection context using these methods.

Key facts:

Given the new paradigm for application development at SAP, we strongly recommend using ABAP-managed database objects for accessing data in SAP Business Integrity Screening.
ABAP-managed database objects are one of the recommended patterns to use in the context of ABAP development on SAP HANA.
Since ABAP Managed Database Procedures (AMDP) are implemented as methods of a global ABAP class, the editing environment for AMDP is the ABAP class editor.
AMDP is written in a database-specific language, such as Native SQL or SQLScript, and is implemented in an AMDP method body of an ABAP class.
The editing if AMDP objects is only supported in the Eclipse-based development environment (ABAP Development Tools) and not in SAP GUI.

Benefits

There is one integrated development environment for application objects and database objects.
The development environment provides syntax checks, code completion etc.
ABAP data types can be used.
Transport- and lifecycle-management is fully integrated into the ABAP environment.
No database user is required for the development and testing.
All native SAP HANA features and objects are available by using SQLScript.

Implementation details

For data access in SAP Business Integrity Screening, the following ABAP-managed database objects are relevant:

Core Data Services (CDS):
- CDS Views: Declarative data modeling language.
- CDS table functions: Use SQLScript, thereby enabling all SAP HANA features and objects.
ABAP Managed Database Procedures (AMDP):
- Implemented in SQLScript.

If you do not use the SAP HANA rules framework, you can do a full migration of all your customer defined Business Content to ABAP-managed database objects (as of SAP Business Integrity Screening 1.3 SP01).

You can also start gradually and do a partial migration of only your Investigation Objects and Detection Objects and Association View, or only your Detection Methods.

Detection methods for Address screening

These methods are based on the Fuzzy-Search engine of the SAP HANA database and used to screen names and addresses maintained in the business partner data or on the transactions against screening lists, provided by governments and authorities (like OFAC).

The address screening methods are available out-of-the-box and it is only required to set up the address screening functionality and set up the screening parameters to use them.

For more information about the address screening functionality in the SAP BIS solution, refer to the chapter "Exploring address screening functionality" of this training material.

Predictive analysis methods

Predictive detection methods are used to identify unknown detection patterns using machine-learning functionality of the SAP HANA database (PAL/APL).

Key facts:

Predictive detection methods use the predictive analysis features of SAP HANA.
Predictive detection methods learn on their own how to classify detection objects as fraudulent or non-fraudulent, compliant or non-compliant, using machine learning algorithms to discover correlations between data and decision.
The higher the score from such a method, the more likely it is that, upon investigation, a detection object will have an alert with confirmed fraud or non-compliance.
When you create a predictive detection method, the system applies a machine learning algorithm to the training view.
The quality of the specific model type is measured by the predictive power and prediction confidence. Both measures take values between 0 and 1. Trained models with both values close to 1 can be trusted to a high degree.
Running a trained model means assigning a predictive detection method to a detection strategy.
For every selected detection object, the model calculates a score. The higher the score, the higher the probability that an investigation would reveal a true positive (confirmed) case.
The score-threshold is a parameter of the detection method. You can lower or raise the score-threshold, using calibration, to find the optimal amount of alerts your organization can handle.

Implementation details

By leveraging Predictive Analytics Integrator (PAI) you can create a classification or regression algorithm directly in the SAP BIS Fiori UI which will automatically result in a Predictive Detection Method.

Predictive Analytics Integrator is an optional SAP HANA Component which provides the feature to create classification or regression models in the SAP Fiori UI:

You need to have a data model in place (Detection and Investigation Objects).

Additionally, you only need a data view and a historical decision view (your classified dataset).

Business end-users can create and update (by retraining) the predictive model in this SAP BIS Fiori UI.

The UI also provides you with an overview about the key influencing variables of the predictive model.

After successful set up and calibration, the predictive detection method can be used for detection of suspicious transactions and master data records.

Next lesson