Sharif University of Technology

Computer Engineering Faculty

 

             Data Mining Using Adaptive Local Searches Based on Meta-Lamarckian Learning

 

By

Maryam Amirhaeri

Zahra Ahmadi

 

 

Under Supervision of

Dr. Jafar Habibi

 

August 2007

 

ABSTRACT

Adaptation of parameters and operators is one of the most important areas of evolutionary computations. In this paper we represent local search algorithms for classification of network attacks using the concept of meta-Lamarckian learning. Here, our interests are on the type of evolutionary algorithms which known as adaptive memetic algorithms. One of unique features of adaptive memetic algorithms is the choice of local search methods or memes and recent studies have shown that this choice has significant effects on the performance of problem searches. Methods which is represented in this paper are based on memetic algorithms to improve the rule set got from genetic cycle. These algorithms are incremental and in each stage of algorithm, a new rule is generated and added to rule set. Using represented adaptive local search algorithms, more on the basis of random selection, there are possibilities to optimize rules created in the genetic cycle. These algorithms are used for classification of network attacks on the DARPA data set, which has information on computer networks, and the results and performance are presented in this paper.

 

KEYWORDS

Adaptation algorithm; Memetic algorithm; Evolutionary algorithm; Genetic algorithm; Optimization; Meta-Lamarckian learning; Intrusion detection