Skip to main content

The Yellow Crane, Inc.

Data Mining

Turning background into the deep understanding of underlying patterns and trends

In The Yellow Crane, Inc. we use several methods and processes for extracting useful patterns and knowledge from large datasets, each with distinct steps and techniques. Here's an overview of some prominent data mining methods along with detailed descriptions and examples of applications:

  • CRISP-DM (Cross-Industry Standard Process for Data Mining)

CRISP-DM is a widely used data mining process model that provides a structured approach to planning and executing data mining projects. It's like following a recipe to ensure you bake the perfect cake, from understanding what kind of cake you want to bake to making sure it tastes good before serving. It consists of six phases:

  • Business Understanding:

Description: Determining the business objectives and converting them into a data mining problem.

Example: An institution wanting to increase sales through targeted marketing. The objective is to identify customer segments that are most likely to respond to marketing campaigns.

  • Data Understanding:

Description: Collecting initial data, understanding the data, and identifying data quality issues.

Example: Gathering customer transaction data, demographic data, and response data from previous campaigns to understand patterns and relationships.

  • Data Preparation:

Description: Cleaning, constructing, and formatting data to be ready for modeling.

Example: Handling missing values, normalizing data, and creating new features such as the frequency of purchases or average transaction value.

  • Modeling:

Description: Selecting and applying various modeling techniques and fine-tune model parameters.

Example: Applying clustering algorithms to segment customers into groups or using classification techniques to predict which customers are likely to respond to a campaign.

  • Evaluation:

Description: Evaluating the models to ensure they meet business objectives.

Example: Assessing model performance using metrics like accuracy, precision, recall, or AUC-ROC, and validating against a hold-out sample to ensure generalizability.

  • Deployment:

Description: Implementing the model in the operational environment.

Example: Integrating the predictive model into the institution’s CRM system to target specific customer segments with tailored marketing offers.

  • SEMMA (Sample, Explore, Modify, Model, Assess)

SEMMA is a methodology developed by SAS Institute, focused on the core tasks of data mining. It involves sampling data, exploring it for patterns, modifying it to enhance models, modeling to identify patterns, and assessing the results. It's like sculpting a statue, where you start with a raw block of marble (data), explore its potential, shape it, refine it, and finally assess the sculpture.

  • Sample:

Description: Extract a portion of the data suitable for analysis.

Example: Selecting a representative sample of customer data for initial exploration.

  • Explore:

Description: Performing exploratory data analysis to discover patterns and anomalies.

Example: Using visualization tools to identify trends in purchase behavior or outliers.

  • Modify:

Description: Transforming and preparing the data for modeling.

Example: Applying data transformations such as normalization, encoding categorical variables, or creating new features based on domain knowledge.

  • Model:

Description: Applying statistical and machine learning models to the data.

Example: Using decision trees, neural networks, or regression analysis to create predictive models for customer churn.

  • Assess:

Description: Evaluating the accuracy and validity of the models.

Example: Validating the model using cross-validation techniques and assessing performance using metrics like F1 score or RMSE.

  • KDD (Knowledge Discovery in Databases)

KDD is a process of discovering useful knowledge from a collection of data. It's like mining for gold, where you need to sift through a lot of dirt (data) to find the valuable nuggets (useful insights). It includes several steps:

  • Selection:

Description: Selecting the relevant data to be analyzed.

Example: Choosing customer transaction data from the past year for analysis.

  • Preprocessing:

Description: Cleaning and preprocessing data to remove noise and handle missing values.

Example: Filling missing demographic information and remove duplicate records.

  • Transformation:

Description: Transforming data into suitable formats for mining.

Example: Aggregating monthly transaction data into quarterly summaries.

  • Data Mining:

Description: Applying data mining techniques to extract patterns.

Example: Using association rule mining to find product combinations frequently purchased together.

  • Interpretation/Evaluation:

Description: Interpret the results and evaluate their significance and usefulness.

Example: Determine the business relevance of the discovered patterns and how they can inform marketing strategies.


Each methodology follows a structured approach to ensure the effective application of data mining techniques. They all emphasize understanding the business problem, preparing and understanding the data, modeling, and validating results, albeit with different terminologies and focus areas. We successfuly applied them across variety of industries, such as finance, retail, healthcare, but primary targeting higher education field. Such diversity is highlighting the versatility and importance of data mining in modern decision-making processes.