Chapter 5 Results and Analysis
5.7 Experimental Results of MSGA
5.7.4 Performance Analysis of the Proposed Method with Different Single
Single Seeds Based Methods
Since an initial population has a significant effect on generating a best population in further generation, single seeds based genetic algorithms generate a smaller number of positive association rules, which are shown in Figures 43-54. Form the experimental results it can be seen that some seeds generate a large number of rules for some muta- tion and crossover operators, but for other crossover operators these seeds generate a smaller number of rules for the same mutation operators. On the other hand, the results obtained by the MSGA present better or similarly high quality rules for different muta-
tion and crossover operators for all data sets than the rules obtained by single seeds based genetic algorithms.
For the Breast Cancer data set, seed 1 and seed 4 have fitness values of 1 and 0.21, re- spectively. According to Figure 43, the seed 1 based genetic algorithm generates a large number of high quality rules using insertion (INS) mutation and uniform crossover op- erators with respect to other single seeds based genetic algorithms. Whereas, seed 4 based genetic algorithm performs better than other seeds based genetic algorithms using displacement (DISP), inversion (INV), scramble (SCM) mutation and uniform crosso- ver operators. From the above analysis it can be concluded that, a genetic algorithm based on a single seed having a high fitness value cannot guarantee that it will generate a large number of high quality rules using different mutation and crossover operators for all data sets. This is also true for a lower fitness value based seed chromosome. On the other hand, MSGA which comprised all seeds to generate an initial population has a significant effect for further generation of best population and this approach present bet- ter or similar high quality rules using different mutation and crossover operators for all data sets, which are shown in Figures 43-54.
Figure 43: Performance of different seeds and MSGA for different mutation operators with uni- form crossover for a Breast Cancer data set.
The performance of different seeds and MSGA for different mutation operators with uniform crossover for a Breast Cancer data set is shown in Figure 43. According to Fig- ure 43, MSGA performs better than other single seed based genetic algorithms for all mutation operators.
Figure 44: Performance of different seeds and MSGA for different mutation operators with single point crossover for a Breast Cancer data set.
The performance of different seeds and MSGA for different mutation operators with single point crossover for a Breast Cancer data set is shown in Figure 44. According to Figure 44, MSGA performs better than or similarly to other single seed based genetic algorithms for all mutation operators.
Figure 45: Performance of different seeds and MSGA for different mutation operators with partial- ly mapped crossover for a Breast Cancer data set.
The performance of different seeds and MSGA for different mutation operators with partially mapped crossover for a Breast Cancer data set is shown in Figure 45. Accord- ing to Figure 45, for all mutation operators the number of rules generated by MSGA is higher than other single seed based genetic algorithms.
Figure 46: Performance of different seeds and MSGA for different mutation operators with uni- form crossover for a Solar Flare data set.
The performance of different seeds and MSGA for different mutation operators with uniform crossover for a Solar Flare data set is shown in Figure 46. According to Figure 46, for all mutation operators the number of generated rules by MSGA is higher than other single seed based genetic algorithms.
Figure 47: Performance of different seeds and MSGA for different mutation operators with single point crossover for a Solar Flare data set.
The performance of different seeds and MSGA for different mutation operators with single point crossover for a Solar Flare data set is shown in Figure 47. According to Figure 47, for all mutation operators the number of generated rules by MSGA is higher than other single seed based genetic algorithms.
Figure 48: Performance of different seeds and MSGA for different mutation operators with partial- ly mapped crossover for a Solar Flare data set.
The performance of different seeds and MSGA for different mutation operators with partially mapped crossover for a Solar Flare data set is shown in Figure 48. According to Figure 48, the number of generated rules by MSGA is higher than other single seed based genetic algorithms for all mutation operators.
Figure 49: Performance of different seeds and MSGA for different mutation operators with uni- form crossover for a Monk’s Problems data set.
The performance of different seeds and MSGA for different mutation operators with uniform crossover for a Monk’s Problems data set is shown in Figure 49. According to Figure 49, for all mutation operators the number of rules generated by MSGA is higher than other single seed based genetic algorithms.
Figure 50: Performance of different seeds and MSGA for different mutation operators with single point crossover for a Monk’s Problems data set.
The performance of different seeds and MSGA for different mutation operators with single point crossover for a Monk’s Problems data set is shown in Figure 50. According to Figure 50, the number of generated rules by MSGA is higher than other single seed based genetic algorithms for all mutation operators.
Figure 51: Performance of different seeds and MSGA for different mutation operators with partial- ly mapped crossover for a Monk’s Problems data set.
The performance of different seeds and MSGA for different mutation operators with partially mapped crossover for a Monk’s Problems data set is shown in Figure 51. Ac- cording to Figure 51, for all mutation operators the number of generated rules by MSGA is higher than other single seed based genetic algorithms.
Figure 52: Performance of different seeds and MSGA for different mutation operators with uni- form crossover for a Mushroom data set.
The performance of different seeds and MSGA for different mutation operators with uniform crossover for a Mushroom data set is shown in Figure 52. According to Figure 52, for all mutation operators the number of rules generated by MSGA is higher than or similarly to other single seed based genetic algorithms.
Figure 53: Performance of different seeds and MSGA for different mutation operators with single point crossover for a Mushroom data set.
The performance of different seeds and MSGA for different mutation operators with single point crossover for a Mushroom data set is shown in Figure 53. According to Figure 53, for all mutation operators the number of generated rules by MSGA is higher than or similar to other single seed based genetic algorithms.
Figure 54: Performance of different seeds and MSGA for different mutation operators with partial- ly mapped crossover for a Mushroom data set.
The performance of different seeds and MSGA for different mutation operators with partially mapped crossover for a Mushroom data set is shown in Figure 54. According to Figure 54, for all mutation operators the number of generated rules by MSGA is higher than or similar to other single seed based genetic algorithms.