An AIS algorithm for Web usage mining with directed mutation

This paper presents a model based on artificial immune system for mining Web usage data. One of the new features of the proposed model is directed mutation that is designed to avoid the random nature of mutation that make the system nondeterministic, besides that the model presents a new method for learning new unseen antigens instead of using the hypermutation which its computational cost is high. In the proposed algorithm each gene in the antigen has its own strength so strong genes are recognized more powerfully. Experimental results show that by exerting the directed mutation and considering item weights in noisy data like Web log data the quality of extracted antibodies are improved and by using the new method for learning new antigens, outliers canpsilat penetrate to set of antibodies. Like the natural immune system, the strongest advantage of immune based learning is its ease of adaptation to the dynamic environment. By introducing the new features, a model which is shown to be more robust and better able to adapt to the dynamic environments such as Web than the similar models is proposed.