Global Optimization Ensemble Model for Classification Methods

Table 2

(a) Parameter optimization using grid search. (b) Parameter configuration for classifier.

(a)  

Parameter	Operator	Grid range	Combination	Scale	Optimal value
Population size	GA-layer 1	2–100	2, 3, 6, 11, 18, 27, 37, 50, 65, 81, 100	Quadratic	6
Maximum number of generation	GA-layer 1	1–50	1, 6, 11, 16, 21, 26, 30, 35, 40, 45, 50	Linear	16
Number of iterations	CV-layer 2	2–50	2, 4, 6, 10, 14, 19, 26, 33, 41, 50	Quadratic	10
Sampling Size	Bagging-layer 3	0–1.0	0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1	Linear	0.6
Number of iterations	Bagging-layer 3	1–100	1, 2, 5, 10, 17, 26, 37, 50, 64, 81, 100	Quadratic	10

(b)  

Operator name	Parameter configuration
ID 3	Criterion: information_gain
	Minimal size of split: 4
	Minimal leaf size: 2
	Minimal gain: 0.1
Decision tree	Criterion: Information_gain
	Minimal size for split: 4
	Minimal leaf size: 2
	Minimal gain: 0.1
	Maximal depth: 20
	Confidence: 0.5
Random forest	Number of trees: 10
	Criterion: Information_gain
	Minimal leaf size: 2
	Minimal gain: 0.1
	Maximum depth: 20
	Confidence: 0.5
Rule induction	Criterion: information_gain
	Sample ratio: 0.7
	Pureness: 0.6
	Minimal prune benefit: 0.6
K-NN	K nearest neighbors: 11
	Weighted vote: true
	Measure type: nominal measures
	Nominal measure: Dice similarity
Naïve bayes	Laplace correction: true
W-AODE	Frequency for super parents: 1.0
W-PART	Confidence threshold: 0.5
	Minimum objects per leaf: 2.0
W-J48	Confidence threshold: 0.5
	Minimum objects per leaf: 2.0