AI Glossary / Data science and analytics / Intermediate

Cross-industry standard process for data mining

Definition

The Cross-Industry Standard Process for Data Mining, usually abbreviated CRISP-DM, is a process model for organizing data-mining projects. Its six phases are business understanding, data understanding, data preparation, modeling, evaluation, and deployment. The model links analytical work to a practical objective and recognizes that teams often move back and forth between phases as they learn more.

What it means

A project does not begin and end with fitting a model. The team must understand the decision it is trying to support, determine whether suitable data exist, prepare those data, assess the result, and decide how the work will be used. CRISP-DM gives those activities a shared structure. The word standard in its name does not make completion of the phases a certification of model quality, legal compliance, or business success.

How it works

Business understanding defines objectives and success criteria. Data understanding examines the available information and its limitations. Data preparation produces the material needed for analysis. Modeling develops candidate analytical approaches. Evaluation checks whether the result adequately addresses the business problem. Deployment puts the outcome into use, which may involve an operational system or an analytical report. The phases are iterative: discovering unsuitable data or an evaluation failure can require revisiting earlier decisions. The process should be adapted to the project rather than followed as a rigid one-way checklist.

Three practical examples

• Customer-service planning: a team clarifies the staffing decision before choosing a forecasting model and data sources.
• Quality analysis: investigators use data understanding to discover that inconsistent defect labels must be resolved before comparing models.
• AI product development: a project connects evaluation criteria to the user's task and plans how a model's results will enter an actual workflow.

Think of it this way

Think of planning and opening a new workshop. Understanding the job, inspecting materials, preparing equipment, testing methods, checking results, and organizing everyday operation are connected activities. Discovering a problem may send the team back to an earlier decision. The analogy illustrates iterative project work; it does not define every technical or governance activity required for a modern AI system.

Worked example

Suppose a fictional business receives 1,000 support cases each week but can manually review only 100. Its objective is to prioritize that review, rather than maximize overall classification accuracy. During business understanding, the team defines success around identifying relevant cases within the review capacity. Data understanding reveals that labels are available only for cases previously selected for review, creating a potential selection problem.
The team revisits the data plan before modeling. Later, evaluation finds that one model identifies 60 relevant cases among its top 100, while another identifies 50. This is useful evidence for the defined prioritization task, but deployment still requires workload procedures, acceptable error tradeoffs, and appropriate monitoring. The example shows why a model score must be interpreted within a complete project decision rather than treated as the final outcome.

Illustrative case study

In a fictional distribution company, analysts are asked to build an AI delay predictor. A CRISP-DM review reveals that dispatchers cannot act on predictions made after the final loading decision. The team redefines the prediction time, removes information unavailable at that moment, and evaluates the revised model using the intended workflow. It then decides how dispatchers will receive and use the output. No performance improvement is assumed; the main result of the scenario is alignment between the analysis and a decision that can actually be made.

Common misconceptions and mistakes

CRISP-DM is a process model, not a particular algorithm or software product. Its phases are not necessarily six separate departments or a fixed schedule. Data preparation is not complete merely because a file can be opened, and evaluation is broader than comparing one technical metric. Deployment is not always an automated production service; an appropriately delivered analytical finding may be the output. Modern AI projects also need explicit attention to matters such as security, privacy, monitoring, and maintenance rather than assuming the phase names fully specify them.

Business use

Businesses can use the framework to make project assumptions, deliverables, and responsibilities visible. At each phase, record the evidence needed to continue and the issues that might require revision. For AI work, connect model evaluation to task outcomes and document operational ownership. A small project can use a lightweight version, while a consequential deployment needs more detailed controls. The framework is useful when it improves decisions and communication, not when teams simply create six folders and declare the methodology complete.

People also ask

Does CRISP-DM require a particular machine-learning method?

No. The modeling choices depend on the objective and data.

Must the phases occur once in order?

No. Iteration is a central feature because new findings can change earlier assumptions.

Does using CRISP-DM guarantee a deployable model?

No. A sound project may conclude that the data, performance, or business conditions do not justify deployment.

Check your understanding

A model performs well using information that becomes available only after the business decision. Which process issue does this reveal?

Answer: The analysis is misaligned with the intended use. The team should revisit business and data understanding, then prepare and evaluate inputs available at the actual decision time.

Related terms

• Data mining
• Deployment
• business understanding
• data preparation
• model evaluation
• data science lifecycle
• MLOps
• success criteria

Learn next

map one small analytical project to the six phases and name a concrete output for each. Identify a finding that would send the team back to an earlier phase. Then add the operational and governance responsibilities needed for the particular use rather than assuming the generic model supplies every detail.

More in Data science and analytics

• Continuous variable - A continuous variable is a numerical variable modeled as capable of taking any value within an interval, rather than only separate countable values.
• Contrast set learning - Contrast set learning searches for combinations of attribute values whose prevalence differs meaningfully across predefined groups.
• Cosine similarity - Cosine similarity measures the alignment of two nonzero numerical vectors by taking the cosine of the angle between them.
• Dark data - Dark data is information an organization has collected or retained but does not meaningfully use for analytics or decision-making.
• Data cleaning - Data cleaning is the process of identifying and addressing errors, inconsistencies, and other quality problems in data so that the data are more suitable...
• Data farming - Data farming is the purposeful generation of data from simulation models through designed experiments.

Sources

Sources: IBM's CRISP-DM documentation describes the process model and iterative lifecycle: https://www.ibm.com/docs/en/spss-modeler/saas?topic=dm-crisp-help-overview . IBM's original process overview illustrates the phase relationships: https://www.ibm.com/analytics/us/en/events/leadership-summit/SALS_Predictive_Maintenance.pdf . Relevant documentation and presentation excerpts were consulted. The support-review numbers and distribution-company scenario are original teaching examples, not measured implementation results.

Previous: Cross-fitting Next: Cross / joint bilateral filter