The framework consists of three complementary modeling views, which together serve to mediate the viewpoints of business people, data scientists, and data engineers. In this page, we illustrates a simplified example of the three modeling views for the banking domain.

Business View

The Business View provides a conceptualization of machine learning requirements. It shows how Business Goals are refined into Decision Goals and Question Goals, and how such Questions can be answered by (machine-learning-generated) Insight elements. The following figure is a simplified example of Business View for banking domain. It shows that a Case worker aims to Process applications and is measured by the Average resolution time metric. As part of processing applications, the case worker needs to make the Decision on credit applications. To make that decison over each application, the case worker needs to know What will be credit risk of current application at hand?. This question casn be answered by a Credit Risk Predictive Model. At runtime, this predictive model receivs an Applicant profile and labels their risk as High or Low per instance.

Analytics Design View

The Analytics Design View represents the design of machine learning solutions for addressing the requirements expressed in the Business View. It models a solution in terms of Algorithms, Softgoals (non-functional requirements), Inluences, and performance Indicators. The following figure is a simplified example of Analytics Design View for banking domain. It shows that Decision Trees and Support Vector Machines (SVM) are among machine learning algorithms that can be used for Classification of applicant profiles. It also shows softgoals such as Interpretability of model that are critical for the system.

Data Preparation View

The Data Preparation View conceptualizes the design of data preparation tasks in terms of data tables, operations, and flows. The following figure is a simplified example of Data Preparation View for banking domain. It shows that Filter and Join operations are used, among others to gerate the Application Profile data table.

Linking the Three Views

These views are linked together to generate a holistic conceptualization of how enterprise strategies are connected to machine learning algorithms and to data preparation activities. The following figure illustrates the three modeling views linked together for the banking domain example. It shows that the Applicant Profile data table is required for performing Classification of applicant profiles which will then generates the Credit Risk Predictive Model.

In the next section, we will cover the language metamodels and definitions. You can also find more examples in the references page.