Top Page | Upper Page | Contents | About This Site | JAPANESE

Label Classification

Label classification is also called "pattern recognition". For computers, image data is the collection of data of the color and the place. So some methods are needed to get information about "line" or "human face" from the data.

By the way, the meaning of "pattern recognition" is used in a larger way. Pattern recognition is the method to judge the answer for "What is this?".

Process of Label Classification

There are two steps in pattern recognition.

• Modeling : The process, giving data along some indexes and models, and memorizing them in the computer , is also called " Machine Learning ". This step is to summarize the information of the pattern.
• Judge the pattern what we want to know : Get information of the most suitable pattern.

Use of Label Classification

There are two ways to use Pattern Recognition.

Cause and Effect Analysis

For this way, we use first step of Pattern Recognition.

If we get the information of the relationship between combination of variances and pattern, it can be the big hint of Cause and Effect Analysis.

Discrimination and Prediction

For this way, we use two steps of Pattern Recognition.

The problem to predict "A or B?" is called "Discriminant Problem".

Kinds of Label Classification

• By Line : Discriminant Analysis , Support Vector Machine , Decision Tree
• By Distance : Discriminant Analysis , MT method , k-NN
• By Function : Logistic Regression Analysis , Neural Network
• By Rule : Rough Sets Analysis , Bayesian Network

Difference of Good Distribution by Methods

Difference of Output by Methods

NEXT Confusion Matrix