SEMMA

SEMMA is a data mining methodology developed by SAS Institute as a framework for organizing the predictive analytics workflow. The acronym stands for Sample, Explore, Modify, Model, and Assess. While closely associated with SAS Enterprise Miner, SEMMA has been described as a general workflow for preparing and evaluating predictive models.

The Sample phase involves selecting a representative subset of data for analysis to balance computational efficiency with statistical validity. Explore focuses on examining data distributions, relationships, quality issues, and potential outliers to inform preprocessing and modeling decisions. Modify covers data preparation and feature engineering, including normalization, encoding, and the creation of new attributes. Model refers to applying predictive modeling techniques—such as decision trees, neural networks, regression, and others—to the prepared data. Assess encompasses evaluating model performance using metrics appropriate to the task (for example, accuracy, precision, recall, AUC) and assessing its business relevance or potential impact.

SEMMA is characterized as a tool-centric workflow, emphasizing practical steps facilitated by data-mining software. It is often contrasted with CRISP-DM, a broader framework that places more emphasis on business understanding and process standardization. Critics note that SEMMA may provide less explicit guidance on problem formulation and deployment compared to some other methodologies, and its phases can be iterative rather than strictly linear.

Despite these considerations, SEMMA remains a widely cited approach for structuring predictive analytics projects, particularly within environments that rely on SAS tools.