First steps – in Data Mining – with Weka

View Github Repository Open presentation in a new window

lkobylinski

See all presentation from lkobylinski

First steps – in Data Mining – with Weka

0 0

codepot-weka

On Github lkobylinski / codepot-weka

First steps

in Data Mining

with Weka

Łukasz Kobyliński & Radosław Szmit

Codepot 2015

What is Data Mining?

Data Mining is a process of discovering hidden information in data.

https://visualisingadvocacy.org/blog/disinformation-visualization-how-lie-datavis

Typical applications

Customer analysis

  • which customers are likely to increase their purchases?
  • which products are more likely to sell to my customers?

Typical applications

Text mining

  • what is the category of this email we have received?
  • is this product review positive or negative?
  • what are they saying about me on twitter?

Typical applications

Image mining

  • which images in my collection contain cats?
  • which of my contacts are visible on these photos?
  • what is the sex and age of these people?

Data Mining Methods

  • Regression analysis
  • Classification
  • Cluster analysis
  • Association rule mining
  • Sequence mining
  • Anomaly detection
  • ...

Task #1

Assign names to flowers

How do they differ?

iris-versicolor, iris-setosa, iris-virginica

iris- iris- iris- iris-setosa ma niskie wartości dla wszystkich parametrów oprócz sepal-width (szerokie działki korony). iris-versicolor ma średnie wartości dla wszsytkich parametrów. iris-virginica ma wysokie wartości dla wszystkich parametrów oprócz sepal-width.

Fisher's iris dataset

petals and sepals

Fisher's iris dataset

http://mbostock.github.io/d3/talk/20111116/iris-splom.html

Task #2

Answer the questions:

  • iris-setosa has:
    • low values of
    • high values of
  • iris-virginica has:
    • low values of
    • high values of
  • the three classes are best separated by:
    • sepallength
    • sepalwidth
    • petalwidth

Task #3

Use the rules.PART classifier on the iris dataset and answer the questions:

  • which is better on this dataset: J48 or PART?
  • how many examples of iris-versicolor have been classified as iris-setosa?
  • how many examples of iris-virginica have been classified as iris-versicolor?
  • what is the accuracy of the classifier on the training set?

Task #4

Use the J48 classifier on the iris dataset and answer the questions:

  • use the tree visualization pane to manually perform classification of the following example:
    • sepallength=6.7, sepalwidth=3.0
    • petallength=5.0, petalwidth=1.7
  • are all the attributes used in the classifier?
  • what are the numbers of instances of type iris-versicolor, which were misclassified as iris-virginica? (use the Visualize classifier errors panel).

Data: the evolution

Big DataData Mining and Knowledge DiscoveryData WarehousingData AccessData Collection

Data: the evolution

Big Data"What’s likely to happen to online sales, considering 1M visits/day?"Data Mining and Knowledge Discovery"What’s likely to happen to Boston unit sales next month? Why?"Data Warehousing"What were unit sales in New England last March? Drill down to Boston."Data Access"What were unit sales in New England last March?"Data Collection"What was my total revenue in the last five years?"

THE END

Thanks!

Łukasz Kobyliński Radosław Szmit http://www.sages.com.pl
First steps in Data Mining with Weka Łukasz Kobyliński & Radosław Szmit Codepot 2015