Page 140 - Control Theory in Biomedical Engineering

124 Control theory in biomedical engineering

Table 9 Comparison analysis with related works.
Reference Classifier Features Output ACC (%)
Lassoued and Ketata MLP 10 2 87.000
(2017)
Lassoued and Ketata MLP 62 5 93.800
(2018a)
Lassoued and Ketata MLP 30 5 93.700
(2018b)
Lassoued and Ketata MLP+GA 30 5 99.999
(2018b)
Lassoued and Ketata MLP 10 2 86.400
(2018c)
Lassoued and Ketata RBF 10 2 99.999
(2018c)
Lassoued and Ketata PNN 10 2 79.500
(2018c)
The proposed work FLC+GA 6 5 97.054
(Genetic FLC)

3.2 Comparison analysis with related works
In order to evaluate the efficacy of this work, a second comparison study
with related works was conducted (see Table 9). It recapitulates the
obtained accuracy by the proposed FLC and other neural network classi-
fiers. These classifiers were previously investigated in other works using
different neural network approaches (Lassoued and Ketata, 2017, 2018a,
b, c). Only the proposed work deals with the fuzzy arrhythmia classifica-
tion. Moreover, each classifier has used its proper methodology mainly
regarding the number of features and classes (outputs). Thus, referring
to Table 9, all the proposed classifiers are effective to about 99.999%–
79.500% of ACC. However, we conclude that despite not achieving
themostaccurateresults (ACC¼99.999%), the proposed FLC is consid-
ered as a precise classifier (ACC¼97.054%). It is considered more inter-
pretable and explicitly defined than the neural network classifiers.

4 Conclusion

In this chapter, we proposed a FLC for classifying arrhythmias. We used
48 recordings from the MIT-BIH Arrhythmia Database. The proposed
fuzzy based system consists of two main blocks: ECG pre-processing and
fuzzy arrhythmia classification. In the first block, the elimination of the