Data Mining Notes

Steps

1. Define the problem;

2. Gather the dataset;

3. Dataset cleaning:

    A. Following the problem, get ride of the unuseful data;

    B. Make up the missing data using the average method or others;

4. Analysis

Analysis Method

1. Feature Extraction Methods

Target: set the rank of the data in order to define which is more important to recognize the class

A. T-test

Target: using t to define the difference of the class. If t is big, the parameter is more useful to distinguish the class.

Steps

a.  Calculate the average (u) of Class 1 and Class 2

b. Using following formula to calculate the s 

c. Using following formula to calculate the t 

Example:

Class 1	Class 2
25	5
16	12
21	9
18	13
32	19

u1 = 1/5*(25+16+21+18+32)=22.4

u2 = 1/5*(5+12+9+13+19)=11.6

     = 40.3

      = 26.8

= 2.948

B. Principle Component Analysis (PCA)

Target: find out the direction efficient for representation.

Example: input (x1, x2) output (u1, u2)

C. Linear Discriminant Analysis (LDA)

Target: Find a linear subspace that maximizes class separability among the feature vector.

Example: input (x1,x2,class) output (u1, u2)

2. Classification

A. Prototype based methods: K-nearest Neighbour (KNN)

B. Boundary based methods: Multiple Layer Perception (MLP)