Search

This paper has proposed a new AIS immune network classifier called Flexible Immune Network Recognition System (FINERS) that uses HVDM as a distance metric for heterogeneous data type without the need for the discretization or transformation of the data into specific type. The experimental results show that the immune network model produces a better accuracy in most of the heterogeneous datasets and it also generates less rules compared to previous immune classification models. Comparing FINERS to FAIRS, although there are no differences in the accuracy for the heterogeneous data, using network feature from the immune system decreases the number of rules in the classifiers. The study solves some limitation shown in (Watkins, 2001; Freitas & Timmis, 2007; Hart & Timmis, 2008; Timmis, 2006). However, FINERS does not show a significant different or improvement on the accuracy and rules reduction on non-heterogeneous data compared to the previous AIS classification models. In conclusion, the results suggest that the use of network feature and to process data in its original types can increase accuracy performance while reducing the number of rules in heterogeneous data. Furthermore, it is significant to process the data in its original types to avoid degradation of data accuracy and it decreases the time in pre processing of data. For the future investigation, other AIS algorithm can employ HVDM function for other tasks such as optimization and clustering. FINERS could also be further refined to make it dynamic and be able to process dynamic data such as time series data. With the result, we hope to derive a more stable and flexible AIS classifier.