Input: Training set |
Output: Hyper-edge set |
Step 1. (Initialize hypergraph) |
FOR each in X DO |
Create one hyper-edge of sample : First, Inherit attributes from the sample randomly |
and replace the values randomly on the rest attribute. Second, inherits the decision |
attribute of . Third, if has missing value, we fill the attribute value in terms with |
continuous attributes or discrete attribute. |
; |
END FOR |
Step 2. (Classify the training set) |
According to formula (7), calculating the neighborhood threshold for each sample; |
FOR each in X DO |
FOR each in E DO |
According to formula (1), calculate the relevant degree of and , . |
IF THEN ; END IF |
END FOR |
FOR each in DO |
IF THEN |
; // is the decision attribute value |
END IF |
END FOR |
Compute the classification of , . |
END FOR |
Compute the correctly classified ratio of the training set: accuracy; |
IF or THEN GOTO Step ; |
ELSE GOTO Step ; |
END IF |
Step 3. (Replace hyper-edge) |
; //the initialize of hyper-edge replacement set |
FOR each in DO |
According to Definition 7 and formula (5), calculate the confidence degree of : . |
IF THEN ; ; END IF |
END FOR |
While we replace the hyper-edge, it is prior to replace the hyper-edge which is generated by |
the sample with missing values. |
WHILE () |
Generate a new hyper-edge through the process similar to Step . |
; ; |
END WHILE |
GOTO Step ; |
Step 4. (Return) |
RETURN E; |