Intelligent systems in accounting, finance and managementassessing predictive performance of ann-based classifiers, страница 19

Laplace. Our first hypothesis (H1) is satisfied (Table XVII), i.e. GA performing better than RT in refining the solution. However, the difference between accuracy rates is not as obvious as it was for the ‘real’ data from experiment 1. This is explainable, since in the latter case (only ‘real’ data) the starting solution has relatively low accuracy rates (80–90%) and it could have been easily improved, whereas in this experiment (centralized data) we have some starting solutions with high accuracy rates (95–98%) that would be difficult to improve whatever the training mechanism used

Table XIII. Three-way ANOVA for training

Source

Type III sum of squares

DF

Mean square

F

Sig.

Partial η2

Model

979 340.512

30

32 644.684

6 322 228.263

0.000

1.000

PREPROC

540.706

2

270.353

52 358.680

0.000

0.999

DISTRIB

148.280

4

37.070

7179.276

0.000

0.997

TECHNIQ

6.396

1

6.396

1238.708

0.000

0.943

PREPROC ×

138.900

8

17.362

3362.559

0.000

0.997

DISTRIB PREPROC ×

9.486

2

4.743

918.554

0.000

0.961

TECHNIQUE DISTRIB ×

1.426

4

0.356

69.036

0.000

0.786

TECHNIQUE PREPROC ×

2.574

8

0.322

62.310

0.000

0.869

DISTRIB ×

TECHNIQUE

Error

0.387

75

0.005

Total

979 340.899

105

Table XIV. 3-

way ANOVA for t

esting

Source

Type III sum of squares

DF

Mean square

F

Sig.

Partial η2

Model

844 172.942

30

28 139.09

506 683.910

0.000

1.000

PREPROC

1296.821

2

648.411

11 675.541

0.000

0.997

DISTRIB

904.441

4

226.110

4071.431

0.000

0.995

TECHNIQUE

3.537

1

3.537

63.695

0.000

0.459

PREPROC ×

605.016

8

75.627

1361.771

0.000

0.993

DISTRIB PREPROC ×

10.714

2

5.357

96.461

0.000

0.720

TECHNIQUE DISTRIB ×

5.432

4

1.358

24.454

0.000

0.566

TECHNIQUE PREPROC ×

10.002

8

1.250

22.511

0.000

0.706

DISTRIB ×

TECHNIQUE

Error

4.165

75

0.056

Total

844 177.107

105

Table XV. Pairs’ comparison for ‘preprocessing’ factor

PREPROCa                         PREPROCa                     Mean difference                 SE                  Sig.                          95% confidence interval