A Comparative Study of GA, PSO and Hybrid PSO-GA for Feature Selection
Intrusion Detection is a technique to identify the abnormal behavior of system due to attack. The unusual behavior of the environment is then identified and steps are taken and methods are formed to classify and recognize attacks. Data set containing a number of records sometimes may decrease the classifiers performance due to redundancy of data. The other problems may include memory requirements and processing power so we need to either reduce the number of data or the number of records. Feature Selection techniques are used to reduce the vertical largeness of data set. This project makes a comparative study of Particle Swarm Optimization, Genetic Algorithm and a hybrid of the two where we see that PSO being simpler swarm algorithm works for feature selection problems but since it is problem dependent and more over its stochastic approach makes it less efficient in terms of error reduction compared to GA. In standard PSO, the non-oscillatory route can quickly cause a particle to stagnate and also it may prematurely converge on sub optimal solutions that are not even guaranteed to be local optimum. A further drawback is that stochastic approaches have problem-dependent performance. This dependency usually results from the parameter settings in each algorithm. The different parameter settings for a stochastic search algorithm result in high performance variances. In this project the modification strategies are proposed in PSO using GA. Experimental results show that GA performs better than PSO for the feature selection in terms of error reduction problems whereas hybrid outperforms both the model in terms of error reduction.
Intrusion Detection is a technique to identify the abnormal behavior of system due to attack. The unusual behavior of the environment is then identified and steps are taken and methods are formed to classify and recognize attacks. Data set containing a number of records sometimes may decrease the classifiers performance due to redundancy of data. The other problems may include memory requirements and processing power so we need to either reduce the number of data or the number of records. Feature Selection techniques are used to reduce the vertical largeness of data set. This project makes a comparative study of Particle Swarm Optimization, Genetic Algorithm and a hybrid of the two where we see that PSO being simpler swarm algorithm works for feature selection problems but since it is problem dependent and more over its stochastic approach makes it less efficient in terms of error reduction compared to GA. In standard PSO, the non-oscillatory route can quickly cause a particle to stagnate and also it may prematurely converge on sub optimal solutions that are not even guaranteed to be local optimum. A further drawback is that stochastic approaches have problem-dependent performance. This dependency usually results from the parameter settings in each algorithm. The different parameter settings for a stochastic search algorithm result in high performance variances. In this project the modification strategies are proposed in PSO using GA. Experimental results show that GA performs better than PSO for the feature selection in terms of error reduction problems whereas hybrid outperforms both the model in terms of error reduction.