The Case against Accuracy Estimation for Comparing Induction Algorithms
ICML '98 Proceedings of the Fifteenth International Conference on Machine Learning
Artificial immune system inspired behavior-based anti-spam filter
Soft Computing - A Fusion of Foundations, Methodologies and Applications - Web intelligence and change discovery
Immune-Inspired adaptive information filtering
ICARIS'06 Proceedings of the 5th international conference on Artificial Immune Systems
Immunity from spam: an analysis of an artificial immune system for junk email detection
ICARIS'05 Proceedings of the 4th international conference on Artificial Immune Systems
A Markov chain model of the b-cell algorithm
ICARIS'05 Proceedings of the 4th international conference on Artificial Immune Systems
| Hi-index | 0.00 |
An existing system - AISEC - which categorises email as interesting or uninteresting using an immune-inspired algorithm is implemented as a plug-in to Outlook to allow seamless user testing. Experiments are performed with a new, large data set to validate previous published results. We show comparable results can be obtained on different data-sets ifthe system parameters are correctly tuned; the algorithm is particularly sensitive to certain parameters. Some flaws in the original algorithm are identified; a modification is proposed to the learning process of the algorithm and to the mutation operator. Tests with the modified algorithm in a number of scenarios in which users' interests frequently change show the improved algorithm is capable of continuously adapting to achieve high classification accuracy and can accurately track changes in user interests. The improvements are statistically significant when compared to the original system.