Abstract
ARTdECOS is a new artificial neural network model, which incorporates fuzzy adaptive resonance theory (ART) at its core and implements evolving connectionist system (ECOS) methodology to enable knowledge extraction from real time, ongoing processes. The model creates category clusters based on resonance threshold conditions with the fuzzy ART core. After a number of input presentations the model amalgamates categories based upon the fuzzy similarity of weight vectors. Four new parameters are introduced to control and evaluate category amalgamation: weight activation threshold, weight resonance threshold, sum of weight differences and minimum number of categories. A graphical visualization tool is presented that depicts category rules and their evolution. Further, a method is presented to normalize input data in real time without setting feature minimum and maximum levels a priori. ARTdECOS is shown to satisfactorily classify instances from the IRIS dataset into three categories in an unsupervised mode. ARTdECOS is implemented for examples of heart rate interval time series. Heart rate variability features are derived based upon fractal and time domains. Segmentation of the heart rate time series data is shown by category selection.
Similar content being viewed by others
Abbreviations
- ECG:
-
Electrocardiogram
- ART:
-
Adaptive resonance theory
- ECOS:
-
Evolving connectionist systems
- ARTdECOS:
-
Adaptive resonance theory with evolving connectionist systems
- HRV:
-
Heart rate variability
- ANN:
-
Artificial neural network
- ANS:
-
Autonomic nervous system
- FuNN:
-
Fuzzy neural network
- FCM:
-
Fuzzy c-means
References
Asuncion A, Newman DJ (2007) UCI machine learning Repository [http://www.ics.uci.edu/~mlearn/MLRepository.html]. University of California, School of Information and Computer Science, Irvine, CA. Accessed on 1 March 2011
Bezdek JC, Ehrlich R, Full W (1984) FCM: the fuzzy c-means clustering algorithm, Comput Geosci 10(2–3):191–203
Carpenter GA, Tan AH (1995) Rule extraction: from neural architecture to symbolic representation. Connect Sci 7(1):3–27
Carpenter GA, Grossberg S, Rosen DB (1991) Fuzzy ART: fast stable learning and categorization of analog patterns by an adaptive resonance system. Neural Netw 4:759–771
Dekker JM, Schouten EG, Klootwijk P, Pool J, Swenne CA, Kromhout D (1997) Heart rate variability from short electrocardiographic recordings predicts mortality from all causes in middle-aged and elderly men. Am J Epidemiol 145(10):899–908
Esteller R, Vachtsevanos G, Echauz J, Litt B (2001) A comparison of waveform fractal dimension algorithms. IEEE Transact Circuits Syst–I: fundamental theory and applications 48(2):177–183
Goldberger Ary L (1996) Non-linear dynamics for clinicians: chaos theory, fractals, and complexity at the bedside. Lancet 347:1312–1314
Hassett AL, Radvanski DC, Vaschillo EG, Vaschillo B, Sigal LH, Karavidas MK, Buyske S, Lehrer PM (2007) A pilot study of the efficacy of heart rate variability (HRV) biofeedback in patients with fibromyalgia. Appl Psychophysiol Biofeedback 32:1–10
Higuchi T (1988) Approach to an irregular time series on the basis of fractal theory. Physica D 31(2):277–283
Isawa H, Matsushita H, Nishio Y (2008) Fuzzy adaptive resonance theory combining overlapping category in consideration of connections. 2008 International Joint Conference on Neural Networks (IJCNN 2008), pp 3594–3599
Jessica DPM, Gevirtz RN, Scher B, Guarneri E (2004) Biofeedback treatment increases heart rate variability in patients with known coronary artery disease. Am Heart J 147(3):G1–G6
Karavidas MK, Lehrer PM, Vaschillo E, Vaschillo B, Marin H, Buyske S, Malinovsky I, Radvanski D, Hassett A (2007) Preliminary results of an open label study of heart rate variability biofeedback for the treatment of major depression. Appl Psychophysiol Biofeedback 32:19–30
Kasabov N (1998) The ECOS framework and the ‘eco’ training method for evolving connectionist systems. J Adv Comput Intell 2(6):1–8
Kasabov Nikola (2009) Integrative connectionist learning systems inspired by nature: current models, future trends and challenges. Nat Comput 8(2):199–218
Kasabov N, Kim JS, Watts M, Gray A (1997) FuNN/2—a fuzzy neural network architecture for adaptive learning and knowledge acquisition. Inf Sci 101(3–4):155–175
Kozma R, Swope JA, Kasabov NK, Williams MJA (1997) Multi-agent implementation of fractal analysis of fuzzy neural networks. Proceedings of the 1997 International Conference on Neural Information Processing and Intelligent Information Systems, pp 162–165
Luczak H, Laurig W (1973) An analysis of heart rate variability. Ergonomics 16:85–97
Nolan RP, Kamath MV, Floras JS, Stanley J, Pang C, Picton P, Young QR (2005) Heart rate variability biofeedback as a behavioral neurocardiac intervention to enhance vagal heart rate control. Am Heart J 149:1137e1–1137e7
Peng C-K, Havlin S, Stanley HE, Goldberger AL (1995) Quantification of scaling exponents and crossover phenomena in nonstationary heartbeat time series. Chaos 5(1):82–87
Reiner R (2008) Integrating a portable biofeedback device into clinical practice for patients with anxiety disorders: results of a pilot study. Appl Psychophysiol Biofeedback 33:55–61
Siepmann M, Aykac V, Unterdörfer J, Petrowski K, Mueck-Weymann M (2008) A pilot study on the effects of heart rate variability biofeedback in patients with depression and in healthy subjects. Appl Psychophysiol Biofeedback 33:195–201
Swope J (2012) Feature selection and adaptive connectionist classification models and a system for biological time series analysis on the case study of heart rate variability. PhD Thesis, Otago University, Dunedin, New Zealand (to be published)
Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology (1996) Heart rate variability. Standards of measurement, physiological interpretation, and clinical use. Eur Heart J 17:354–381
Zhang L, Wang G, Wang W (2006) A new fuzzy ART neural network based on dual competition and resonance technique. In: Wang J, Yi Z, Zurad JM, Lu BL, Yin H (eds) Advances in neural networks—ISNN 2006, Third International Symposium on Neural Networks, Chengdu, China, Proceedings Part I, Springer, Berlin, pp 792–797
Author information
Authors and Affiliations
Corresponding author
Appendix ARTdECOS process steps
Appendix ARTdECOS process steps
-
A.
Model preparation
-
1.
Allocate memory space for the maximum allowable number of weight vectors, each having length of two times the maximum number of features.
-
2.
Initialize all weight vector components to 1.
-
3.
Set the active number of category nodes to 1.
-
4.
Initialize the model parameters. Suggested values are:
$$ \rho = 0. 8 2\quad \beta = 0. 7 6\quad \alpha = 0. 1\quad cn_{ \min } = 1\quad A_{W} = 0. 7 5\quad \rho_{W} = 0. 6 5 $$ -
5.
Decide upon the method of choosing scaling limits.
-
1.
-
B.
Input presentation for each instance
-
1.
If adaptive scaling is used:
-
a.
Determine if new instance is outside current limits.
-
b.
Adjust scaling if necessary, suggested β S = 0.9.
-
c.
Adjust existing weights if necessary.
-
d.
Record new scaling limits.
-
a.
-
2.
Normalize input vector in [0 1].
-
3.
Determine complement of each input vector component.
-
4.
Form new input vector from steps 2 and 3.
-
5.
Determine activation and resonance of input vector with each category.
-
6.
Find category having maximum activation and satisfying resonance condition.
-
7.
If no category satisfies resonance condition, then create a new category.
-
8.
Adapt winning category to current instance.
-
9.
Record category selected for current instance.
-
1.
-
C.
Category amalgamation
-
1.
Determine if sufficient instances have been presented to begin amalgamation.
-
a.
This may be a set length of data or a particular dataset.
-
b.
Or, the maximum number of categories has been reached.
-
a.
-
2.
Determine the pair of weights having the highest activation relative to each other.
-
3.
If the pair of weights in step two exceeds weight activation threshold and weight resonance threshold, then merge these two weights. If either threshold is not met, then stop here.
-
4.
Reassign to the newly formed category index the winning category for each instance previously assigned to either merging category.
-
5.
Decide upon further amalgamation.
-
a.
If amalgamation has been triggered by the maximum number of categories having been reached (step 1b), then stop here.
-
b.
If amalgamation is at a set length of data or a particular dataset (step 1a), then go back to step 2 unless the minimum set categories has been reached.
-
a.
-
6.
If the input data used in the modelling is available, then present the data to the amalgamated set of categories for reassignment. This is done without any additional learning in the weight vectors.
Rights and permissions
About this article
Cite this article
Swope, J.A. ARTdECOS, adaptive evolving connectionist model and application to heart rate variability. Evolving Systems 3, 95–109 (2012). https://doi.org/10.1007/s12530-012-9049-2
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s12530-012-9049-2