Objectives of the Analytics Design View

The Analytics Design View aims to support exploration of alternate approaches for the analytical and machine learning problem at hand. It looks to facilitate design of (machine learning) experiments and identify trade-offs. It supports algorithm selection and monitoring their performance over time.

Understanding the elements of the Analytics Design View Model

In Analytics Design View, the main modeling elements are Analytics Goals, Algorithms, Softgoals, and Influences. The table below breaks down what each element represents, how they are applied, and their importance.

Type of Element Description
Analytics Goals Analytics Goals capture the intention of the analysis to be performed over the datasets. Three types of Analytics Goals are distinguished. The type of Analytics Goal can be derived from the type of Insight that is required to generate (from the Business View). Each Analytics Goal is then connected to its corresponding Insight element via the generates link.
Algorithms Algorithms are procedures and calculation steps that are needed to fulfill an Analytics Goal. They are connected to Analytics Goals through the performs link, showing a means-end relationship.
Softgoals Softgoals represent quality requirements to be taken into account during design of the machine learning solution.
Influences Influence Links show how the Softgoals are satisfied through operationalization and design decisions. This view is connected to the previous modeling view through the generates link.


Analytics Goals are broken down into further types : Prediction Goal, Description Goal, and Prescription Goal. The table below breaks down these attributes.

Type of Analytics Goal Description
Prediction Goal If the analytics aims to predict the value of a data attribute (i.e., a variable or data column), it is called a Prediction Goal.
Description Goal If the analytics aims to summarize and explain the dataset, it is called a Description Goal.
Prescription Goal If the analytics aims to find the optimal alternative given a set of options and criteria, it is called a Prescription Goal.


Constructing the Analytics Design View Model

Refer to the following links to follow a step-by-step methodology of constructing Analytics Design View models:

Step 1: Specify the top level Analytics Goals that the system would achieve
Step 2: Decompose Analytics Goals
Step 3: Model a set of Algorithms that can fulfill your top-level Analytics Goals
Step 4: Model the criteria for making design decisions and algorithm selection in terms of Softgoals and Indicators
Step 5: Model the Influence Links from Algorithms to Softgoals
Step 6: Model the Influence Links from Algorithms to Indicators


Example of Analytics Design View Modelling in Practice


Below is an example illustrating the connection between an Analytics Design View Model and the Business and Data Preparation Views. The middle-part shows parts of the Analytics Design View developed in this example along with links to the Business View (top-part) and Data Preparation View (bottom-part).

Business View - Physician