SEMMA

Sample · Explore · Modify · Model · Assess

1. Overview of SEMMA — SAS’s Standard Data Mining Process, From Sampling Large Datasets to Model Assessment

    flowchart LR
    A["Large-scale<br/>source data"] --"Sample (S) →<br/>Explore (E) → Modify (M)"--> B["Analysis-ready dataset<br/>and predictive model"] --"Assess (A)<br/>business validation"--> C["Apply data mining<br/>results"]

    style A fill:#E3F2FD,stroke:#1976D2,color:#000
    style B fill:#FFF3E0,stroke:#F57C00,color:#000
    style C fill:#E8F5E9,stroke:#388E3C,color:#000

Definition: A standard data mining process proposed by SAS Institute that cycles through five stages — Sample, Explore, Modify, Model, and Assess — on large datasets to develop and validate predictive models.

Characteristics: (Focused on technical procedure) Centers on the technical procedure of data processing — the business-understanding stage is treated as a separate, preceding step. (Repeatable structure) Each stage can be repeated independently, maximizing analytical efficiency in large-scale data environments. (Tight SAS integration) Closely integrated with the SAS Enterprise Miner tool, and widely applied in practical analytics environments.

2. Core Structure of SEMMA

A. The Five-Stage Analysis Process

    flowchart LR
    S["Sample<br/>Extract a<br/>representative<br/>sample"]
    E["Explore<br/>Statistics/visualization<br/>Identify outliers/patterns"]
    M1["Modify<br/>Transform/create variables<br/>Handle missing values/outliers"]
    M2["Model<br/>Select algorithm<br/>Train/validate model"]
    A["Assess<br/>Measure performance<br/>Review business value"]

    S --> E --> M1 --> M2 --> A
    A -->|"Rerun if performance<br/>falls short"| S

    style S  fill:#E3F2FD,stroke:#1976D2,color:#000
    style E  fill:#F3E5F5,stroke:#7B1FA2,color:#000
    style M1 fill:#FFF3E0,stroke:#F57C00,color:#000
    style M2 fill:#FFEBEE,stroke:#D32F2F,color:#000
    style A  fill:#E8F5E9,stroke:#388E3C,color:#000

Stage	Core Purpose	Key Activities	Key Techniques/Tools
S — Sample	Extracting a representative sample for analytical efficiency	Stratified sampling, random sampling, over-/under-sampling	Random sampling, SMOTE, stratified sampling
E — Explore	Understanding data distribution, patterns, and outliers	EDA (exploratory data analysis), correlation analysis, visualization	Histograms, scatter plots, correlation matrices
M — Modify	Transforming data into a form optimized for analysis	Imputing missing values, handling outliers, feature engineering, normalization	Standardization, one-hot encoding, derived-variable creation
M — Model	Training and optimizing a model for the prediction goal	Algorithm selection, cross-validation, hyperparameter tuning	Regression, decision trees, SVM, ensembles
A — Assess	Final evaluation of model performance and business value	Measuring accuracy/AUC/F1, business-impact analysis	ROC curve, lift chart, profit matrix

B. Comparison with CRISP-DM and Application Guidance

    flowchart TD
    subgraph R1["　"]
        direction LR
        C1["SEMMA<br/>Proposed by SAS Institute<br/>Focused on technical data-<br/>processing steps<br/>5 stages (S-E-M-M-A)"]
        C2["CRISP-DM<br/>An industry-standard process<br/>Includes business context<br/>6 stages (BU-DU-DP-M-E-D)"]
    end
    subgraph R2["　"]
        direction LR
        C3["Common ground<br/>Iterative structure<br/>Data prep, modeling,<br/>and evaluation cycle"]
        C4["Selection criteria<br/>SAS environment: SEMMA fits<br/>Cross-team collaboration:<br/>CRISP-DM<br/>Industry standard needed:<br/>CRISP-DM"]
    end

    style C1 fill:#E3F2FD,stroke:#1976D2,color:#000
    style C2 fill:#F3E5F5,stroke:#7B1FA2,color:#000
    style C3 fill:#FFF3E0,stroke:#F57C00,color:#000
    style C4 fill:#E8F5E9,stroke:#388E3C,color:#000
    style R1 fill:none,stroke:none
    style R2 fill:none,stroke:none

Comparison	SEMMA	CRISP-DM
Proposed by	SAS Institute	An industry consortium including SPSS, Daimler, and NCR
Number of stages	5 stages (S-E-M-M-A)	6 stages (BU-DU-DP-M-E-D)
Business understanding	A separate, preceding step (not included)	Explicitly included as stage 1
Deployment stage	Not included	Explicitly included as Deployment
Tool integration	Optimized for SAS Enterprise Miner	Tool-independent
Best-fit environment	SAS-based analytics environments, technically focused projects	Business-data team collaboration, enterprise-wide analytics standardization

3. Expected Benefits and Application of the SEMMA Methodology

Category	Key Benefit	Application in Practice
Structured analysis	Standardizing the stages of a data mining project	Customer-churn prediction and credit-scoring model development in finance, telecom, and retail
Quality assurance	Improved model reliability through thorough data preparation in the Explore/Modify stages	Resolving class-imbalance issues through optimized sampling strategy
Iterative improvement	Gradually improving model performance through evaluation-driven cycles	Tracing underperforming models back to their root-cause stage and rerunning
Leveraging the SAS ecosystem	Seamless process integration with SAS Enterprise Miner	Building automated analytics pipelines in SAS-based organizations

CRISP-DM Graph Theory Framework