Real-World Applications - Immunological Computation: Theory and Applications - page 189

wherein an antibody represents schedule, whereas an antigen holds information on
the set of expected arrival dates for each job to the shop. A library of antibodies
builds new solutions to the problem.
Ong et al. (2005), in a similar work called “ClonaFLEX,” intends to solve the
fl exible JSSP with recirculation. It employs self-initiated antibody initialization,
suitable antibody mutation rates based on their a nities, and a novel distribution
of elite pools to produce antibodies. h e possible job schedules are modeled as
antibodies. h e search process is repeatedly carried out and the information gained
in each generation is used as feedback to conserve and propagate good features.
ClonaFLEX also employs parallel search for optimizing time.
7.7 AIS in Data Mining
Timmis and Knight (2001) wrote a chapter on the concepts of artifi cial immune
systems, particularly on artifi cial immune networks (AINE), which is a machine-
learning algorithm, based on immune network theory as applicable to the fi eld of
data mining. h e self-organizing nature of B cell network can be used as an e -
cient clustering tool. h e idea of repertoire completeness is achieved by making a
certain receptor surface match not only to an exact complementary string, but also
to some variations of it, that is, a ball of recognition. h e B cell receptors can serve
as cluster centers, which will suitably self-organize.
Hunt and Cooke (1996) attempted to apply an immune-based model to data
mining by creating a system that could help in the customer-profi ling domain.
Each B cell object contained customer profi le data such as marital status, owner-
ship of cars, and bank account details.
Serapião et al. (2007) used an artifi cial immune system for the classifi cation of
petroleum well drilling operations. Particularly, two approaches based on CLON-
ALG and parallel AIRS2 were developed. h ey implemented a system, which takes
advantage of information collected by mud-logging techniques during well-drilling
operations. Mud-logging systems operations collect two types of information: forma-
tion samples (shale-shaker samples) and mechanical parameters related to the drill-
ing operation. AIRS2 is a bone marrow clonal selection type of immune algorithm.
AIRS2, as in CLONALG, develops a set of memory cells that represents the training
data environment. Also, AIRS2 uses a nity maturation and somatic hypermutation.
It works on two stages: evolving candidate memory cells and determining whether
they should be added to the pool of memory cells or not. Once the training routine is
performed, AIRS2 classifi es instances using k -nearest neighbor ( k -NN) on the set of
developed memory cells. h us, AIRS2 fi rst learns the input space through a cluster-
ing process and then uses k -NN on the cluster representatives for classifi cation. h e
reported results showed that imbalanced real mud-logging data has large impact on
the classifi cation performance of the AIS classifi ers; they achieve high precision on
predominant classes, but lower classifi cation precision on classes with fewer samples.