introduction to data mining and machine learning techniques · introduction to data mining and...

Introduction to Data Mining and MachineLearning TechniquesIza Moise, Evangelos Pournaras, Dirk Helbing

Iza Moise, Evangelos Pournaras, Dirk Helbing 1

Overview

Main principles of data miningDefinitionSteps of a data mining processSupervised vs. unsupervised data mining

Applications

Data mining functionalities

Iza Moise, Evangelos Pournaras, Dirk Helbing 2

Definition

Data mining

is the automated process of discovering interesting (non-trivial, pre-viously unknown, insightful and potentially useful) information orpatterns, as well as descriptive, understandable, and predictive mod-els from large-scale data.

Also known as:

• Knowledge discovery in databases (KDD)

• Data analysis

• Information harvesting

• Business intelligence

Iza Moise, Evangelos Pournaras, Dirk Helbing 3

Goals

X search consistent patterns and/or systemic relationshipsbetween data

X validate the findings by applying the detected patterns to newsubsets of data

Iza Moise, Evangelos Pournaras, Dirk Helbing 4

Data Mining is...

• k-means clustering

• decision trees

• neural networks

• Bayesian networks

Iza Moise, Evangelos Pournaras, Dirk Helbing 5

Data Mining is not...

• Data warehousing

• SQL / Ad Hoc Queries / Reporting

• Software Agents

• Online Analytical Processing (OLAP)

• Data Visualization

Iza Moise, Evangelos Pournaras, Dirk Helbing 6

Knowledge Discovery Process

Iza Moise, Evangelos Pournaras, Dirk Helbing 7

Steps of a data mining process

1. Exploration

2. Model building and validation

3. Deployment

Iza Moise, Evangelos Pournaras, Dirk Helbing 8

Exploration

• data preparation→ data cleaning, data transformation etc.

• elaborate exploratory analyses using a wide variety ofgraphical and statistical methods, depending on the nature ofthe analytic problem

• determine the general nature of models that can be taken intoaccount in the next stage

Iza Moise, Evangelos Pournaras, Dirk Helbing 9

Model building and validation, and Deployment

• Model building and validation:→ consider various models and choose the best one→ validate the model on existing data

• Deployment:→ apply the model to new data

Iza Moise, Evangelos Pournaras, Dirk Helbing 10

Model building and validation, and Deployment

• Model building and validation:→ consider various models and choose the best one→ validate the model on existing data

• Deployment:→ apply the model to new data

Iza Moise, Evangelos Pournaras, Dirk Helbing 10

Model building and validation, and Deployment

• Model building and validation:→ consider various models and choose the best one→ validate the model on existing data

• Deployment:→ apply the model to new data

Iza Moise, Evangelos Pournaras, Dirk Helbing 10

Two styles of data mining: Predictive andDescriptive

1. Predictive→ Predict the value of a specific attribute (target or dependent)based on the value of other attributes (explanatory)

2. Descriptive→ Derive patterns that summarise the relationships betweendata

Iza Moise, Evangelos Pournaras, Dirk Helbing 11

Supervised vs. Unsupervised

Supervised

• predictive or directed

• useful when you have a specifictarget value to predict aboutyour data

• Classification, Regression,Anomaly Detection

Unsupervised

• descriptive or undirected

• finds hidden structure andrelation within the data

• Clustering, Association, FeatureExtraction

Iza Moise, Evangelos Pournaras, Dirk Helbing 12

Supervised Data Mining

• a pre-specified target variable

• explanatory variables ↔ one (or more) dependent variables

• the algorithm is given many training data where the target isknown

• the algorithm identifies the “new” data points that match themodel of each target value

Iza Moise, Evangelos Pournaras, Dirk Helbing 13

Unsupervised Data Mining

• determine the existence of classes or clusters in the data

• exploratory analysis

• all variable are treated in the same way

Iza Moise, Evangelos Pournaras, Dirk Helbing 14

Overview

Main principles of data miningDefinitionSteps of a data mining processSupervised vs. unsupervised data mining

Applications

Data mining functionalities

Iza Moise, Evangelos Pournaras, Dirk Helbing 15

Fielded Applications

• Web mining→ PageRank, Search Engine, Recommenders, Social Networks

• Screening Images

• Diagnosis• Marketing and Sales

→ Market basket analysis, Direct marketing

• Biology, biomedicine, astronomy, chemistry

Iza Moise, Evangelos Pournaras, Dirk Helbing 16

Beers and Diapers

Iza Moise, Evangelos Pournaras, Dirk Helbing 17

Overview

Main principles of data miningDefinitionSteps of a data mining processSupervised vs. unsupervised data mining

Applications

Data mining functionalities

Iza Moise, Evangelos Pournaras, Dirk Helbing 18

1. Supervised Data MiningX ClassificationX RegressionX Outlier detectionX Frequent pattern mining

2. Unsupervised Data MiningX ClusteringX Feature Extraction

→ definition

→ real use-cases

→ method

→ pros and cons

Iza Moise, Evangelos Pournaras, Dirk Helbing 19

1. Supervised Data MiningX ClassificationX RegressionX Outlier detectionX Frequent pattern mining

2. Unsupervised Data MiningX ClusteringX Feature Extraction

→ definition

→ real use-cases

→ method

→ pros and cons

Iza Moise, Evangelos Pournaras, Dirk Helbing 19