|
|
| population_size = 21 |
| //Crossover rate |
| //Mutation rate |
| divide data into 10 equally-sized subsets |
| for cv_round = 1 to 10 do |
| divide training data D- into train (6 subsets) and test (3 subsets) data |
| initialize methods with train data: |
| cwk-NN and wk-NN OVA: HVDM initialization |
| ONE: fitness value calculation for values of attributes |
| evaluate starting population Weights with test data and ONE/cwk-NN/wk-NN OVA |
| while ending terms of GA are not fulfilled do |
| //Survivor selection: Elitism |
| search for the individual with the highest fitness rate from the population |
| //Parent selection: Roulette-wheel selection with fitness-proportionate selection |
| for each individual in the population do |
| calculate individual’s fitness proportionate rate = individual‘s fitness rate/sum of individuals’ |
| fitness rates |
| calculate individual’s cumulative fitness proportionate rate |
| end for |
| while nr of individuals in the mating pool is smaller than population_size do |
| generate a random number r from |
| search for the jth individual that has smaller cumulative fitness proportionate rate than |
| add the jth individual in the mating pool |
| end while |
| //Crossover: Uniform crossover with discrete recombination |
| for each individual in the mating pool do |
| generate a random number s from |
| if is smaller than then |
| add the individual in the parent pool |
| else |
| add the individual in the new population (offspring is a direct copy of its parent) |
| end if |
| end for |
| while two individuals can be taken from the parent pool do |
| if two individuals are exactly the same then |
| add the first individual into the new population |
| take new individual from the parent pool to use in the crossover |
| end if |
| for each disease class weight set do |
| select the crossover points randomly |
| swap information of two individuals in the corresponding crossover points (create children) |
| end for |
| add children in the new population |
| end while |
| //Mutation: Uniform mutation |
| for each individual in the new population do |
| for each gene of individual do |
| generate a random number from |
| if is smaller than then |
| select a random value from the range |
| set the value v as a new value of the gene |
| end if |
| end for |
| end for |
|
evaluate children and mutated individuals in the new population with test data and |
|
ONE/-NN/-NN |
|
add the elite individual without changes into the new population |
|
//Survivor Selection |
|
if nr of individuals in the new population is larger than population_size then |
|
sort cases descending by their fitness rate |
|
discard the last cases in order to have correct nr of individuals in the population |
|
else if nr of individuals in the new population is smaller than population_size then |
|
select randomly missing cases from the old population |
|
end if |
|
end while |
|
initialize methods with training data D-: |
|
cwk-NN and wk-NN OVA: HVDM initialization |
|
ONE: fitness value calculation for values of attributes |
|
evaluate the individual with the highest fitness rate after GA with testing data and |
|
ONE/cwk-NN/wk-NN OVA |
| end for |
