Modelscape Develop

Modelscape™ Develop™ is a set of model development and documentation tools. These tools are intended for risk managers, analysts, and quants who develop, test, and document models for risk assessment and decision support.

Modelscape Develop comprises these tools:

• Interactive apps for data preparation, model construction, model testing, and validation

• A comprehensive library of validation statistics, machine and deep learning, financial, risk, and economic algorithms

• Customizable and reusable model development templates

• Automated model documentation generation

You can use Modelscape Develop to perform these tasks:

• Build, test, experiment with, and validate multiple models in parallel.

• Create models from validated pre-built functions instead of writing code or building your own libraries.

• Automate iterative workflows through code generation and reuse.

• Integrate algorithms and internal IP developed in any language or application.

• Generate live, auditable, and traceable documentation for model validation and governance.

• Preserve model development history for auditability, transparency, and knowledge transfer.

Modelscape Develop Workflow

Use Modelscape Develop to develop statistical and machine learning models in MATLAB^®.

This figure shows how to use the Modelscape Develop workflow in parallel with the Modelscape Validate™ workflow. The Develop workflow comprises the white boxes and the Validate workflow comprises the orange boxes. You can also perform the validation workflow independently after the development workflow.

Preprocess Data Using Live Tasks

Load data for your models in MATLAB. You can preprocess the data using the Remove Risk Factors and Screen Risk Factors live tasks.

Use Remove Risk Factors to interactively inspect variables from a data table and filter them out. You can also add the reasons for including or excluding variables and use the live task to document your analysis. For more information on how to do this, see Remove Risk Factors.

Use the Screen Risk Factors live task to interactively use predefined, customizable screening criteria to filter out input variables based on their predictive power. You can also add the reasons for including or excluding variables and use the live task to document your analysis. For more information on how to do this, see Screen Risk Factors by Custom Criteria.

You can also use the suite of metrics in Modelscape to analyze the bias in your data set. For more details, see Fairness Metrics in Modelscape.

Train Models Using MATLAB

After preprocessing your input features, you can use MATLAB to train your machine learning models. For example, you can use the Classification Learner and Regression Learner apps to train your models.

You can also use the suite of metrics in Modelscape to analyze the bias in your models. For more details, see Fairness Metrics in Modelscape.

Check Performance of Models

You can check the performance of your model using validation metrics available in Modelscape. For more details, see Modelscape Validate. After you compare models, you can select a model that suits your needs.

Perform Model Comparisons Using Experiment Manager

You can compare the performance of your models against each other or against existing models using the app. For more information, see Model Development and Experiment Manager.

Document Model Development Process

Document the results and analyses of the model in a Microsoft^® Word document from MATLAB. Many workflows in financial institutions involve writing and submitting reports to internal control functions or regulatory bodies. These documents often conform to a given house style and are typically Microsoft Word documents. Using Modelscape, you can add text, visualization, and tabular content to a Microsoft Word document from MATLAB. For more information, see Model Documentation in Modelscape.

After you develop a model, you can pass it to one of these stages:

• Validation stage — For more information, see Modelscape Validate.

• Test stage — For more information, see Modelscape Test.

• Deployment stage — For more information, see Modelscape Deploy.