CRISP-DM

Cross-Industry Standard Process for Data Mining

1. Overview: An International Standard Process for Data Analysis to Solve Business Problems

    flowchart LR
    A["Define the<br/>business problem"] --"A 6-step<br/>analysis process"--> B["Data-driven<br/>model development"] --"Iterative<br/>execution"--> C["Create and deploy<br/>business value"]

    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 process for data mining and analytics projects, applicable across industries, that iterates through six stages — from business understanding to deployment — to create business value from data.

Characteristics: (Industry neutrality) A cross-industry standard process that is not tied to any specific tool or technology. (Iterative improvement) Progressive quality improvement through feedback and iteration between stages, rather than a strictly one-directional flow. (Common language) A common language between data analysts and business stakeholders that improves collaboration efficiency.

2. Core Components of CRISP-DM

A. The Six-Stage Analysis Process

    flowchart LR
    S1["Stage 1<br/>Business<br/>Understanding<br/>Define objectives & plan"]
    S2["Stage 2<br/>Data<br/>Understanding<br/>Collect, explore, assess quality"]
    S3["Stage 3<br/>Data<br/>Preparation<br/>Clean, transform, integrate"]
    S4["Stage 4<br/>Modeling<br/>Select algorithm<br/>Train model"]
    S5["Stage 5<br/>Evaluation<br/>Assess performance<br/>Review against objectives"]
    S6["Stage 6<br/>Deployment<br/>Put into production<br/>Monitor"]

    S1 --> S2 --> S3 --> S4 --> S5 --> S6

    style S1 fill:#E3F2FD,stroke:#1976D2,color:#000
    style S2 fill:#E8EAF6,stroke:#3949AB,color:#000
    style S3 fill:#F3E5F5,stroke:#7B1FA2,color:#000
    style S4 fill:#FFF3E0,stroke:#F57C00,color:#000
    style S5 fill:#FFEBEE,stroke:#D32F2F,color:#000
    style S6 fill:#E8F5E9,stroke:#388E3C,color:#000

Stage	Key Activities	Core Deliverables
1. Business Understanding	Define project purpose/scope, set data mining objectives	Project charter, success criteria document
2. Data Understanding	Initial data collection, EDA (exploratory data analysis), quality assessment	Data exploration report, list of quality issues
3. Data Preparation	Handle missing values, remove outliers, feature engineering, data integration	Final analysis-ready dataset
4. Modeling	Select model, split train/validation data, tune hyperparameters	Optimized predictive model
5. Evaluation	Measure model performance metrics, review business value, approve deployment	Model evaluation report, deployment decision document
6. Deployment	Apply to production, train users, build monitoring system	Deployed model, operations manual

B. The Iteration Mechanism

    flowchart TD
    subgraph ROW1["　"]
        direction LR
        BU["Business<br/>Understanding"] --> DU["Data<br/>Understanding"] --> DP["Data<br/>Preparation"]
    end
    subgraph ROW2["　"]
        direction LR
        MOD["Modeling"] --> EVAL["Evaluation"] --> DEP["Deployment"]
    end

    DP --> MOD
    EVAL -->|"Redefine if objectives unmet"| BU
    DP -->|"Quality issue found"| DU

    style BU   fill:#E3F2FD,stroke:#1976D2,color:#000
    style DU   fill:#E8EAF6,stroke:#3949AB,color:#000
    style DP   fill:#F3E5F5,stroke:#7B1FA2,color:#000
    style MOD  fill:#FFF3E0,stroke:#F57C00,color:#000
    style EVAL fill:#FFEBEE,stroke:#D32F2F,color:#000
    style DEP  fill:#E8F5E9,stroke:#388E3C,color:#000
    style ROW1 fill:none,stroke:none
    style ROW2 fill:none,stroke:none

Type of Iteration	When It Occurs	Purpose
Within-stage iteration	Parameter tuning during modeling, adjusting transformation approach during data preparation	Improve the quality of that stage’s output
Feedback to a prior stage	Data preparation → data understanding, modeling → data preparation	Address gaps found in an earlier stage
Whole-project iteration	Re-running from stage 1 when evaluation shows business objectives are unmet	Redefine the analytical direction/hypothesis and rerun the whole process

3. Expected Benefits and Practical Application of CRISP-DM

Category	Key Expected Benefit	Practical Application
Project structuring	Improved predictability and management efficiency of analytics projects	Building a WBS for AI/ML projects and per-stage deliverable management standards
Stronger collaboration	A common language between business and data teams	Stakeholder engagement and expectation alignment through per-stage review meetings
Quality assurance	Improved model reliability through iterative verification	Running a pre-deployment gate that checks alignment with business objectives
Knowledge reuse	Accumulating analytics know-how through a standardized process	Building reusable data pipelines/feature libraries for similar future projects

Data Quality Management (DQC)SEMMA