Data Mining

Data Mining

Data mining - methods and algorithms to explore and analyze large volumes of data

Goal: to find patterns in data that are

• valid: with some certainty
  • e.g. everybody speaks English in Blois - not true
• novel: non obvious for a human
  • everybody speaks French in Blois - obvious
• useful: can do something with extracted knowledge
• understandable for humans

What is DM

What is NOT Data Mining:

• look up a phone number in a dictionary
• compute the number of customers who bought iPad in August
• can use SQL for that

What is Data Mining:

• What is the profile of the customers who bought iPad?
• Which customers will buy the new iPhone?
• Which customers will buy which products?

Origins

DM is a discipline with roots from

• Artificial Intelligence
• Statistics
• Machine Learning
• Pattern Recognition
• Cognitive Science
• Database Systems

Main Focuses

DM is mostly used

• Customer Relationship Management (CRM)
  • churn scoring - predict if a customer leaves to a competitor
  • direct marketing - show ads only to whose who are interested
  • credit scoring
  • sales forecasting
  • etc
• website/search optimization
• supply chain optimization
• many others

Types of Data Mining

Rule Mining

Local Pattern Discovery

• Frequent Pattern Mining
  • Apriori and Eclat algorithms for that
• Association Rule Mining
• Constraint-Based Pattern Mining

Sequence Mining:

• Sequential Pattern Mining

Graph Mining

• Social Network Mining

Others

• Cluster Analysis
• Web Mining
• Text Mining - part of Natural Language Processing and Information Retrieval
• Stream Mining
• Tree Mining
• Preference Mining

Data Mining Process

CRISP-DM (CRoss Industry Standard Process for Data Mining)

Business Understanding

• Define the success criteria
• How to integrate the output with existing technologies?

Data Understanding

• Collect the data from Data Sources
• Summarizing Data: First Look at the Data
• Exploratory Data Analysis
• Univariate Analysis - to analyze how variable values behave in isolation
• Bivariate Analysis - to analyze how two variables interact

Data Preparation

• Need to prepare data so it can be processed by Models
• Data Cleaning
• Data Transformation
• Data Reduction

Data Modeling

• Multivariate Linear Regression
• Logistic Regression
• Decision Tree (Data Mining)
• SVM
• many others

Evaluation

• Error Analysis
• Error Metrics
• Cross-Validation
• Learning Curves
• ROC Analysis

Links

• http://en.wikipedia.org/wiki/Data_mining
• nice DM&ML slides link
• Data Mining syllabus in Boston College link
• Data Mining map by Saed Sayad link

Source

• Data Mining (UFRT)