SAS Certified Specialist Machine Learning Using SAS Viya Quick Facts (2025)

Exam Domain Breakdown

Domain 1: Data Sources (32% of the exam)

Create a project in Model Studio

• Bring data into Model Studio for analysis — Import data from a local source (Import tab).
• Bring data into Model Studio for analysis — Add data from a stored data source (Data Sources tab).
• Bring data into Model Studio for analysis — Use an in-memory data source (Available tab).
• Create Model Studio Pipelines with the New Pipeline window — Automatically generate pipelines.
• Create Model Studio Pipelines with the New Pipeline window — Pipeline templates.
• Advanced Advisor options — Maximum class level.
• Advanced Advisor options — Maximum % missing.
• Advanced Advisor options — Interval cut-off.
• Partition data into training, validation, and test — Explain why partitioning is important.
• Partition data into training, validation, and test — Explain the different methods to partition data (stratified vs simple random).
• Use Event Based Sampling for rare events.
• Set up Node Configuration.

Section summary: In this part of the exam, you will focus on creating and configuring projects within Model Studio. Key tasks include importing data from multiple sources, understanding how to work with in-memory environments, and managing pipeline templates that guide analytical workflows. You will need to know how to establish project parameters and apply advisor settings that influence modeling and data preparation.

You will also explore data partitioning techniques that ensure models are developed and validated on balanced datasets. This includes distinguishing between stratified and simple random partitions, applying event-based sampling for rare targets, and setting up project configurations that optimize performance as you progress through the modeling process.

Explore the data

• Use the DATA EXPLORATION node.
• Profile data during data definition.
• Preliminary data exploration using the data tab.
• Save data with the SAVE DATA node.

Section summary: This section emphasizes exploratory data analysis and profiling. You will use Model Studio’s visual tools to examine distributions, evaluate missing values, and identify variable types that influence model structure. The DATA EXPLORATION node helps you generate statistical summaries that allow you to better understand patterns and relationships within your dataset.

You will also learn how to save processed data using appropriate nodes, ensuring that exploratory steps are preserved for ongoing use. This workflow reinforces best practices in data management and prepares you to transition from exploration into transformation with clean, well-understood inputs.

Modify data

• Explain concepts of replacement, transformation, imputation, filtering, outlier detection.
• Modify metadata within the DATA tab.
• Modify metadata with the MANAGE VARIABLES node.
• Use the REPLACEMENT node to update variable values.
• Use the TRANSFORMATION node to correct problems with input data sources, such as variables distribution or outliers.
• Use the IMPUTE node to impute missing values and create missing value indicators.
• Prepare text data for modeling with the TEXT MINING node.
• Explain common data challenges and remedies for supervised learning.

Section summary: This section covers preparing your dataset for analysis through variable management and correction steps. You will apply nodes for transformation, imputation, and replacement to create consistent input structures that support accurate modeling. Understanding how to detect and remedy data issues such as skewed distributions, missing values, and outliers will be an essential competency.

Additionally, you will explore the TEXT MINING node for formatting and tokenizing textual data to feed into machine learning pipelines. Topics emphasize how thoughtful preprocessing improves supervised learning results, ensuring inputs reflect meaningful patterns rather than noise or inconsistencies.

Use the VARIABLE SELECTION node to identify important variables to be included in a predictive model

• Unsupervised Selection.
• Fast Supervised Selection.
• Linear Regression Selection.
• Decision Tree Selection.
• Forest Selection.
• Gradient Boosting Selection.
• Create Validation from Training.
• Use multiple methods within the same VARIABLE SELECTION node.

Section summary: You will gain proficiency in selecting key input variables that significantly influence model outcomes. The VARIABLE SELECTION node provides a range of methods, from unsupervised approaches to supervised algorithms that focus on prediction strength. You will compare techniques like regression-based selection, tree-based feature identification, and boosted frameworks for refined subset creation.

In addition to understanding when and why each method is applied, you will learn to interpret outputs and validation results. This helps ensure that the predictive models you create are lean, focused, and optimized for performance and interpretability.

Domain 2: Building Models (42% of the exam)

Describe key machine learning terms and concepts

• Data partitioning: training, validation, test data sets.
• Observations (cases), independent (input) variables/features, dependent (target) variables.
• Measurement scales: Interval, ordinal, nominal (categorical), binary variables.
• Supervised vs unsupervised learning.
• Prediction types: decisions, rankings, estimates.
• Curse of dimensionality, redundancy, irrelevancy.
• Decision trees, neural networks, regression models, support vector machines (SVM).
• Model optimization, overfitting, underfitting, model selection.
• Describe ensemble models.
• Explain autotuning.

Section summary: You will review foundational machine learning terminology and principles. This includes understanding how data is split, what kinds of variables are used, and how prediction outputs differ by type. The section also highlights distinctions between supervised and unsupervised approaches, offering a conceptual framework for how each applies to analytical tasks.

The section further explores model optimization and selection strategies, the balance between bias and variance, and how ensemble methods can combine different learners for stronger performance. You will also study autotuning and its role in efficiently identifying optimal hyperparameters using SAS Viya automation.

Build models with decision trees and ensemble of trees

• Explain how decision trees identify split points — Split search algorithm.
• Explain how decision trees identify split points — Recursive partitioning.
• Explain how decision trees identify split points — Decision tree algorithms.
• Explain how decision trees identify split points — Multiway vs. binary splits.
• Explain how decision trees identify split points — Impurity reduction.
• Explain how decision trees identify split points — Gini, entropy, Bonferroni, IGR, FTEST, variance, chi-square, CHAID.
• Explain how decision trees identify split points — Compare methods to grow decision trees for categorical vs continuous response variables.
• Explain the effect of missing values on decision trees.
• Explain surrogate rules.
• Explain the purpose of pruning decision trees.
• Explain bagging vs. boosting methods.
• Build models with the DECISION TREE node — Adjust splitting options.
• Build models with the DECISION TREE node — Adjust pruning options.
• Build models with the GRADIENT BOOSTING node — Adjust general options: number of trees, learning rate, L1/L2 regularization.
• Build models with the GRADIENT BOOSTING node — Adjust Tree Splitting options.
• Build models with the GRADIENT BOOSTING node — Adjust early stopping.
• Build models with the FOREST node — Adjust number of trees.
• Build models with the FOREST node — Adjust tree splitting options.
• Interpret decision tree, gradient boosting, and forest results (fit statistics, output, tree diagrams, tree maps, variable importance, error plots, autotuned results).

Section summary: Here, you will dive deep into tree-based algorithms that underpin much of modern machine learning. You will learn how SAS Viya implements decision trees, forests, and gradient boosting models, and how to tune key parameters for optimal performance. A major focus is placed on understanding splitting criteria, pruning strategies, and surrogate rules for handling missing values automatically.

Interpretation plays a key role, and this section ensures you can read fit statistics and graphical diagnostics, like tree maps and variable importance plots. You will demonstrate your ability to fine-tune iterative tree models and to identify where bagging and boosting improve robustness against variance and bias.

Build models with neural networks

• Describe the characteristics of neural network models — Universal approximation.
• Describe the characteristics of neural network models — Neurons, hidden layers, perceptrons, multilayer perceptrons.
• Describe the characteristics of neural network models — Weights and bias.
• Describe the characteristics of neural network models — Activation functions.
• Describe the characteristics of neural network models — Optimization Methods (LBFGS and Stochastic Gradient Descent).
• Describe the characteristics of neural network models — Variable standardization.
• Describe the characteristics of neural network models — Learning rate, annealing rate, L1/L2 regularization.
• Build models with the NEURAL NETWORK node — Adjust number of layers and neurons.
• Build models with the NEURAL NETWORK node — Adjust optimization options and early stopping criterion.
• Interpret NEURAL NETWORK node results (network diagram, iteration plots, and output).

Section summary: This section focuses on the design and training of neural networks using SAS Viya’s interface. You will better understand how architectures are structured with layers and activation functions, and how weights, learning rates, and regularization techniques influence generalization performance. You will configure these parameters within the NEURAL NETWORK node to create efficient and stable models.

The interpretation of results, such as convergence patterns and iteration plots, is also central. You will learn to explain model complexity and visualize internal relationships in a way that communicates value to both technical and business audiences.

Build models with support vector machines

• Describe the characteristics of support vector machines.
• Build model with the SVM node — Adjust general properties (Kernel, Penalty, Tolerance).
• Interpret SVM node results (Output).

Section summary: You will gain insights into how support vector machines construct decision boundaries by maximizing margins. The section trains you to choose appropriate kernel types, adjust penalty and tolerance levels, and use the SVM node to implement these models effectively within Model Studio.

Detailed review of output results helps ensure you can interpret decision functions and evaluate model quality. This supports a confident understanding of when to apply SVMs versus other machine learning algorithms in SAS Viya.

Use Model Interpretability tools to explain black box models

• Partial Dependence plots.
• Individual Conditional Expectation plots.
• Local Interpretable Model-Agnostic Explanations plots.
• Kernel-SHAP plots.

Section summary: Model interpretability tools play a vital role in explaining predictive behavior. You will explore multiple visualization and interpretive methods for both global and local explanations, ensuring that even complex models can be clearly understood and communicated.

You will learn how to use each technique effectively within SAS Viya to explain model results, balancing predictive accuracy with transparency. The focus is on building trust in machine learning outcomes for both analysts and stakeholders.

Incorporate externally written code

• Open Source Code node.
• SAS Code node.
• Score Code Import node.

Section summary: This section addresses how SAS Viya integrates with open-source languages and custom code workflows. You will learn to incorporate Python, R, or SAS scripts directly into Model Studio pipelines through dedicated nodes, enabling flexibility in modeling and feature engineering approaches.

The section also highlights how imported score code can streamline deployment and unify different programming ecosystems. It’s all about maintaining interoperability while using SAS Viya as the central hub for machine learning execution.

Domain 3: Model Assessment and Deployment (26% of the exam)

Explain the principles of Model Assessment

• Explain different dimensions for model comparison — Training speed.
• Explain different dimensions for model comparison — Model application speed.
• Explain different dimensions for model comparison — Tolerance.
• Explain different dimensions for model comparison — Model clarity.
• Explain honest assessment — Evaluate a model with a holdout data set.
• Use the appropriate fit statistic for different prediction types — Average error for estimates.
• Use the appropriate fit statistic for different prediction types — Misclassification for decisions.
• Explain results from the INSIGHTS tab.

Section summary: This section strengthens your ability to critically evaluate model performance. You will assess metrics such as training efficiency, tolerance, interpretability, and predictive fidelity. Learning when to choose certain fit statistics ensures your evaluations are aligned with business and analytical goals.

The focus extends to building honest assessments using holdout datasets and correctly interpreting results through the INSIGHTS tab. This provides the foundation for consistent, transparent model evaluation within the SAS Viya ecosystem.

Assess and compare models in Model Studio

• Compare models with the MODEL COMPARISON node.
• Compare models with the PIPELINE COMPARISON tab.
• Interpret Fit Statistics, Lift Reports, ROC reports, Event Classification chart.
• Interpret Fairness and Bias plots.

Section summary: This section deals with comparing competing models to identify the best performing approach. You will practice using the MODEL COMPARISON node and pipeline-level analysis features to evaluate statistical performance side-by-side, interpreting key reports such as ROC and Lift charts.

You will also explore fairness and bias evaluation plots, vital for ensuring your models maintain ethical and balanced predictions. The section reinforces how SAS Viya equips practitioners to select trustworthy, data-driven solutions for deployment.

Deploy a model

• Exporting score code.
• Registering a model.
• Publish a model.
• SCORE DATA node.

Section summary: The final section highlights the deployment life cycle within SAS Viya. You will learn to export and register models, generate score code, and publish productions ready for operational environments. The SCORE DATA node plays a key role in executing models efficiently on new inputs.

Topics emphasize the importance of integrating model deployment with corporate governance and automation infrastructure. This ensures that models created in SAS Viya can consistently deliver value, driving real-world insights from data.