You are on page 1of 7

A Simple Demo of Memetic Algorithm Feature Selection

Step 1: Run MAFS and open the GUI java jar MAFS.jar Step 2: Load Data from a file in the format of arff (or CSV, C45). (Here the preprocessing panel of Weka is applied, for more information about the data format and Weka, please refer to http://www.cs.waikato.ac.nz/ml/weka/) Click the button Open file , select a data file in the open dialogue window.

Step 3: Switch to the Memetic Feature Selection panel.

Step 4: Configure the fitness evaluator by clicking the AttributeSubsetEvaluator tab.

Select a classifier (e.g. IB1) on the pop up dialogue window:

We can set the number of folds for cross validation, (defaulted -1, leave-one-out cross validation):

Step 5: Set the parameter of Memetic Algorithm by clicking the tab of MemeticSearch.

In the popping up dialogue window we can set the crossover probability, mutation probability, maximum generation, population size, local search ect. The meaning of each parameter is given as follows: crossoverProb: crossover probability crossoverType: select the type of crossover localSearch: whether to use local search localSearchMethod: select the local search method FilterRankingLS (WFFSA) or MarkovBlanketLS (MBEGA). localSearchInterval: set how many best chromosomes to undergo local search, e.g. if it is set to 3, then the best 3 chromosomes in the population are selected to use local search. maxEvaluation: the maximum allowable number of fitness evaluation. maxGenerations: the maximum allowable number of generation. maxSelectedAttr (minSelectedAttr): fix the max (min) number of bit 1 in each chromosome. mutationProb: mutation probability populationSize: population size, (should be odd number) reportFrequency: the frequency to report the population, it is calculated by generation.

Step 6: Select attribute selection mode: training, cross-validation or bootstrap.

Step 7: Start the search

Read the results in the output text area.

Or you can save the results by right clicking the item in the results lists and select save result buffer

You might also like