@Imtiyaz,

Firstly, I can suggest you normalisation of all fitness factors.

Then you ccan define the weighted fitness function, e.g.

F(Indv) = SNR * w_0 + FalseAlarm * w_1 + (no_examples/all_data) * w_2.

Where w_x values are in a range [-1.0..0.0..1.0] determines how important is given factor. Moreover, in this way you can define min- or maximization problem.

I think this simple approach gives you a good start to your research... and define more specialised fitness function later. The "Multi-Objective Optimization" suggested by [@Ramya mm] is a very good hint.

Your question is vague. Please provide more information about the project you are doing. What do you mean by best match?

each problem has its own fitness function, it is application dependent, that why @Behrouz is asking for more information about your project.

Indeed a GA fitness function is problem dependant. Typically for classification tasks where learning is supervised, error measures ( euclidian, manhatan ...) have been widely adopted. You can also use entropy based approaches as a fitness function.

For optimization problems, you can use a crude model of the function you are investigating.

I think you can try some similarity measures in pattern recognition

This is a simple example..

Http://www.clarkfrancis.com/docs/f_ga.html

@Behrouz

As my project is related to selecting best optimal threshold for detection purpose, and variables which I am choosing for it are

I. SNR

II. Probability of false alarm

III. Number of samples of received data for detection.

But i m stuck with fitness function because no where it is defined how to prepare that. All research article deal only with basic fundamentals of GA and final result, no one tells about how they prepare fitness function. So can u please elaborate, how should I proceed for above defined problem.

@Sreenivas

Link only describe the basics of GA but not how to make fitness function for particular problem...

Hi Imtiyaz,

As everyone mentioned earlier, to write a fitness function it is necessary to know the detailed description of the problem.

I will put in with a simple example. There are 3 variables x,y and z. The problem is to find the best values of x y and z so that their total value is under a threshold t. We have another criterion which states that the sum of x and y should be less than z.

So we create a population of x,y and z. We calculate the fitness as x+y+z

@ Imtiyaz,

As Srinvias said, you would have a condition that decides on whether your threshold is optimal or not.

With each threshold in your population, your GA would come up with an appropriate SNR, Probability of false alarm and no. of sample received. If you are trying to maximize all these parameters then your fitness function would be SNR + false alarm +samples. The sum should be greater than a particular threshold.

Sometimes if there are contradicting objectives, such as you want to maximize SNR and minimize the false alarm, then your objective function would be SNR - false alarm

@ Ramya

do have any reference for the points which you have mentioned regarding

SNR +false alarm+ samples

SNR-false alarm

Selection Fitness function for GA is based on your problem. Either you are working in unsupervised problem of supervised problem. If you explain your problem of work then there is chance to specify the fitness function

@Imtiyaz,

I guess the following link might help

http://www.ise.rutgers.edu/resource/research_paper/paper_05-008.pdf

@ Murty

As I have already defined my problem in one of my comments. Still I m not clear with the definition of fitness function for my problem.

@Imtiyaz,

Firstly, I can suggest you normalisation of all fitness factors.

Then you ccan define the weighted fitness function, e.g.

F(Indv) = SNR * w_0 + FalseAlarm * w_1 + (no_examples/all_data) * w_2.

Where w_x values are in a range [-1.0..0.0..1.0] determines how important is given factor. Moreover, in this way you can define min- or maximization problem.

I think this simple approach gives you a good start to your research... and define more specialised fitness function later. The "Multi-Objective Optimization" suggested by [@Ramya mm] is a very good hint.

GA is a optimization tool, so generally fitness function is a max/min value function consisting of all the variables. As I understand about your problem, you want to find best optimal threshold (i.e. min value of the fitness function), now you have to generate a function with these parameters - SNR, Probability of false alarm and Number of samples of received data for detection in such a way that the value of the function must be tends to zero. This function is called fitness function and the final value of this function after performing GA will be the optimal outcome. The creation of the function is totally depends on your approach towards the solution of the project.

Presumably you will be using a GA to perform an optimization task. How do you measure achieved/estimated performance in your task? Answering this is more or less equivalent to identifying an appropriate fitness function.

GA is simply a search algorithm.. from the pool of population..

As you are selecting attribute.. you must be selecting the Subset from the set of attributes..

The fitness function in this case will the closeness relation, you want to measure with the set...

It may be specific to you desired goal from the selected attributes...

if its general then, we can use many relevance functions like...

Mutual Information

Information Gain

RelieF

Chi-Squre

Correlation

etc..

This is my paper about genetic algorithm in Java, using JGAP library:

http://computationalnonlinear.asmedigitalcollection.asme.org/article.aspx?articleid=1694126

It is a maximization function

Purpose

•Parent selection

•Measure for convergence

•For Steady state: Selection of individuals to die

Khan, You must create an equation where the left side is the value of your fitness function, and the right side uses the attributes to compute the value. The best individual produces the lowest (or highest) value (of a fitness fnc).

Take a look at this R package.

https://cran.r-project.org/web/packages/gaselect/gaselect.pdf

I found it clear and useful for the evaluation of the prediction power of variable subsets. In particular it calculates the fitness of the OLS, PLS, regression models and LM. Hope this can help you.

how about fitness function for genetic algorithm? i will use it for intrusion detection system 

thank

My answer was valid for genetic algorithm.

 you are correct to say that Fitness function is part of genetic algorithm. the truth is, multi objective optimization in genetic algorithm is impossible when you cannot generatte the exact fitness function for the input variables and the corresponding response variable.The best way  to generate the fitness function is to do design of experiment to produce the randomized inputs, then fix in your responses and perform numerical optimization to generate the optimal equation. the optimal equation will be linear if you employ 2-level factorial for the design, it can also be a second order polynomial equation when you employ complete composite design also known as the CCD. once you generate the optimal equation then what is left is to rewrite it in a way that MATLAB will understand it, for example: 2x1 + 3x2 can be written as 2*x(1) + 3*x(2) where x1 and x2 represent two different input variables, for more enquiry you can call me on +2348038027260, or email [email protected]

according to problem ,have to define fintness fuction

Really, fitness function is problem dependent but the values obtained can be btw negative to positive depending on whether you are minimizing or maximizing d objective function.

If it is a classification problem(detection) then accuracy is one of the option for fitness function.

https://towardsdatascience.com/how-to-define-a-fitness-function-in-a-genetic-algorithm-be572b9ea3b4

The objective function evaluates the quality of each solution. It determines the way in which the search strategy is conducted

So even if the problem just requires the difference between an ideal value and your tests values (basically an error function) can be used as ff? I said this because im looking to minimize the difference between my optimization design.

If someone could tell me some tips i'd appreciate it a lot

The fitness function should define the most imprtant parts of the problema so that its optimal value means the solution of the problem

Some times this OF can be a black box

Dear Imtiyaz Khan , One Idea is that, if your data can be organized like a vector with many componants that are independant or othogonal, a fitness function could be the euclidian distance between the estimation (from GA individuals) and the measurements that are the real data.

In this way you will have to minimize this euclidian distance. The best predictor will be the GA individual with the smallest euclidian distance.

It is the objective function that models the problema to be solved

Refer the conference paper it may useful:

Conference Paper Fitness Function of Genetic Algorithm in Structural Constrai...

Basic example in Matlab:

https://www.mathworks.com/help/gads/examples/coding-and-minimizing-a-fitness-function-using-the-genetic-algorithm.html

Some times, it can generally defined in case of an error.

it can defined of an error (IAE,ISE,ITAE.ITSE)

Article Optimal PID controller of a brushless DC motor using genetic algorithm