Data mining exercise Clustering Lab 3

1
Data mining exerciseClusteringLab 3

• Winnie Lam
• Email cswinnie_at_comp.polyu.edu.hk
• Website http//www.comp.polyu.edu.hk/cswinnie/
• The Hong Kong Polytechnic University
• Department of Computing

2
REVIEW

• Classification modeling
• WEKA (ID3)
• Clementine (C5.0)
• Questions?

3
OVERVIEW

• KDD Process

Evaluation
Data Mining
Transformation
Preprocessing
Knowledge
Selection
Patterns
Transformed Data
Preprocessed Data
Target Data
4
Simplified process
Define target discover useful data
Data Understanding
Obtain Clean Useful data
Data Preparation
Discover patterns
Data Mining
Apply the knowledge
Evaluation
5
Download files

• WEKA
• http//prdownloads.sourceforge.net/weka/weka-3-4-8
  a.exe
• Data file
• MyData_lab3.mdb
• http//www.comp.polyu.edu.hk/cswinnie/data/MyData
  _lab3.mdb
• lab3.csv
• http//www.comp.polyu.edu.hk/cswinnie/data/lab3.c
  sv

6
Classification
With predefined class!
7
Clustering
No class is defined previously!
STAR
CROSS
TRIANGLE
8
Clementine

• Clustering with

9
Modeling Tools Clustering

• K-means. An approach to clustering that defines k
  clusters and iteratively assigns records to
  clusters based on distances from the mean of each
  cluster until a stable solution is found.
• TwoSteps. A clustering method that involves
  preclustering the records into a large number of
  subclusters and then applying a hierarchical
  clustering technique to those subclusters to
  define the final clusters.
• Kohonen Networks. A type of neural network used
  for clustering. Also known as a self organizing
  map (SOM).

10
Data Understanding
Data file MyData_lab3.mdb
Step 1 Create Data Source (ODBC) in Control Panel
5
1
7
6
2
3
4
11
Data Understanding
Step 2 Import Data to Clementine

• Add Source node (Database) to Clementine
• Choose Data Source
• Select Tables (lab3, Link, Shop_Info) lt- one at
  each time

1
2
3
4
12
(No Transcript)
13
Data Preparation

14
Goal Merge table lab3 and Shop_Info
Add Node Merge (in Record Ops Palette)
link TID SHOP_CD
lab3 TID dt gp1 gp2 ref_no cl prod_cd
Shop_Info dist_cd shop_cd staffs manager Area
Answer by yourself What is/are the key(s) for
merging?
15

• Step 1
• Merge table lab3 and link

Add Node Merge (in Record Ops Palette)
Field from .Link
16
Step 2. Merge result in step 1 to table Shop_Info
Add Node Merge (in Record Ops Palette)
Fields from .Shop_Info
17
Goal Add a new attribute Weekday and Hour
Useful Node Derive (in Field Ops Palette)
Newly derived
Result
Weekday datetime_day_name(datetime_weekday(dt))
Hour datetime_hour(datetime_time(dt))
18
Discretization
Goal Divide the Hour field into 3 intervals
(Fixed-width)
2
3
1

• Steps
• Add Binning node and specify no. of bins
• Add Type node to update the information of
  newly added information
• Add Re-classify to rename the bins to Morning,
  Afternoon, Evening

19
Data Transformation
Goal Divide the Staff field into 5 intervals
(Fixed-width)
Result
20
Data Mining - Clustering
Goal Divide the customers into 3 clusters
Add Node K-means (in Modeling Palette)
4
1
5
2
3
21
Data Mining - Clustering
Find out unsuitable attributes that cannot
represent the cluster
Result
Note You may adjust the value of k (no. of
clusters)
22
WEKA

• Clustering with

23
Import data lab3.csv
Goal remove useless attributes
1
2
24
Remaining attributes
25
Clustering - SimpleKMeans
Choose Classifier (wekagtclusterersgtSimpleKMeans)
1
2
numClusters -- set number of clusters

seed -- random number seed
26
Result
27
Comparison
Clementine
Weka
28
SUMMARY

• Today, youve learnt
• Derive new attributes
• Merge tables
• Perform discretization (binning)
• Clustering modeling with
• Clementine and WEKA