| Let population be denoted as |
| Probability of crossover be denoted as |
| Probability of mutation be denoted as |
| Fitness function be denoted as |
| Pareto fronts be denoted as |
| |
| repeat |
| Step 1. . |
| Step 2. Initialize population . |
| Initialize number of chromosomes as random individuals which are encoded |
| as strings of 0's and 1's in the chromosome. The length of chromosome depends |
| on the total number of non-zero eigen vectors having non-zero eigen values. |
| {1 indicates inclusion of the eigen vector in the covariance |
| matrix and 0 represents discard of the eigen vector.} |
| Step 3. Evaluate fitness function = number of 1's in the chromosome string. |
| Step 4. Evaluate where is the original matrix and is the transformation |
| matrix. {Based on eigen values selected, map the feature matrix to lower |
| dimension by multiplying the original matrix with the transformation matrix. |
| Step 5. Evaluate fitness = misclassification error rate of the classifier taking |
| B as input matrix.} |
| Step 6. Considering and , perform non-dominated sorting using NSGA-II( ) |
| and generate pareto fronts such as . |
| Step 7. Calculate the crowding distance of all solution points using the crowding distance( ). |
| Step 8. Perform tournament selection by selecting random pairs from . |
| Step 9. Use the crowded comparison operator( ) to select the most widely spread solutions. |
| Step 10. Perform pairwise crossover and bitwise mutation to create new offspring. |
| Step 11. Let the new population be denoted as . |
| until ( of ) |
