Intelligent Signal Analysis Using Case-Based Reasoning for Decision Support in Stress Management - Computational Intelligence in Healthcare: Advanced Methodologies - page 168

Point-to-Point Matching. Finger temperature varies from person to person. FT lies
for some people from 22 0 C to 23 0 C and for others from 34 0 C to 35 0 C. The variation
of FT can be seen in Fig. 18 where 39 measurements are plotted in one chart. The
indication of a person's stress might be found both from low or high temperatures.
Point-to-point matching is not significant in identifying similarities or differences
between two case features because: i) It needs to calculate the difference between all
the measurement points of the two compared cases, which in our study means calcu-
lating the difference for 1800 samples contained in one case. The computational time
is an important issue here, and with several hundred cases in a case library it would
take too long to be of practical use. ii) It does not calculate similarity in patterns of
two cases. Let's consider this example: one case with 15 minute sample data where
finger temperature lies between 33 0 C and 35 0 C, with the same pattern compared
with another case where the finger temperature lies between 24 0 C and 28 0 C and also
follows a similar pattern. From a clinical viewpoint these two cases are similar since
they follow a similar pattern; however Point-to-Point matching would classify them as
not similar according to their different temperature values. iii) If the finger tempera-
ture for a new problem case lies at a higher level (e.g. 35 0 C) then the system will
never consider any cases where the finger temperature lies at a lower level (e.g. 23 0 C)
and vice versa. iv) This similarity method does not consider any variation in FT when
calculating the difference between two cases, whereas the change in the finger tem-
perature is an important factor for classifying stress.
Fig. 18. Sample of FT measurements plotted for 39 sensor signals
Fourier Transform. Discrete Fourier Transform (DFT) and Fast Fourier Transform
(FFT) are efficient and well known methods in Fourier analysis to compute a
sequence of values into components of different frequencies. It is widely employed in
signal processing to analyze the frequencies contained in a sampled signal such as
human voice or speech sounds. However, for the finger temperature sensor signal the
amplitude of the frequency components is not an indication of stress, rather the