Stochastic Behaviour of Parameter Convergence in Genetic Algorithm: An Experimental Analysis

–Parameter convergence in Genetic Algorithm (GA) is quiet unpredictable since it is not tied with the number of episodes (i.e., the number of generations), convergence time, or any other parameter it require to converge. Thus, in such a situation, trial to frame out any hypothesis to predict definite number of episodes gets disapproved; in turn, the alternate hypothesis becomes acceptable that proves that there is non-dependent relationship between various parameters with regard to number of episodes for proper convergence of a Genetic Algorithm. The reason behind is that, it is an optimization depended technique that can take any number of episodes to converge within the given convex set until convergence criteria is satisfied. In other words, unless it converges down to adequate optima, it can go on acquiring more number of episodes. It is to note that, optimization techniques may not provide the pin-pointed target but in most of the cases it guarantee to provide optimum result, i.e., in the proximity of target. Here, an experimental study is undertaken to test the hypothesis in favour of verifying inter-relationship of parameters and its effect on convergence. The GA has been designed for the purpose of generating various parameters of interest to assist the study. Keywords––GA, Genetic Algorithm, Convergence, GA parameters, Episodes, Hypothesis Test,