Intelligent Signal Analysis Using Case-Based Reasoning for Decision Support in Stress Management - Computational Intelligence in Healthcare: Advanced Methodologies - page 170

The decision support system shown in Fig. 20 assists in the case-based classifica-
tion of FT sensor signals in stress management. It takes a finger temperature meas-
urement during the calibration phase and identifies a number of individual parame-
ters. Then, from this sensor signal, it identifies the essential features and formulates a
new problem case with the extracted features in a vector. This new problem case is
then fed into the CBR cycle to retrieve the most similar cases.

Calibration Data

Case formulation

Feature extraction

CBR CYCLE

Similarity Matching
Algorithms
-Fuzzy similarity
-Similarity matrix
-Distance function

Retrieved
cases

Case
library

Reuse

Retain/
learned case

Revise/verified
case

Suggested solution

Confirmed solution

Fig. 20. Case-Based Reasoning for Decision Support in Stress Management

Classification of the FT measurement is done using case retrieval and matching
based on the extracted features. Here, the k-Nearest Neighbour (kNN) algorithm is
applied for the retrieval of similar cases. The new problem case is matched using
different matching algorithms including modified distance function, similarity matrix
and fuzzy similarity matching . A modified distance function uses Euclidean distance
to calculate the distance between the features of two cases. Hence, all the symbolic
features are converted into numeric values before calculating the distance. For exam-
ple, for the feature 'gender', male is converted to one (1) and female is two (2). The
function ' similarity matrix ' is represented as a table where the similarity value be-
tween two features is determined by the domain expert. For example, the similarity
between same genders is defined as 1 otherwise 0.5.
In fuzzy similarity, a triangular membership function ( mf ) replaces the crisp value
of the features for new and old cases with a membership grade of 1. In both cases, the
width of the membership function is fuzzified by 50% on each side. For example, an
attribute ' S' of a current case and an old case have the values -6.3 and -10.9 respec-
tively. If the weight of the membership function ( mf ) is fuzzified with 50 % on each
side, which can be done using a trial and error process based on the problem domain,
the input value for the current case -6.3 can be represented with a mf grade of 1 and
the lower and upper bounds -9.45 and -3.15 can be represented with an mf of grade 0.