Exploring novel features and decision rules to identify cardiovascular autonomic neuropathy using a hybrid of wrapper-filter based feature selection
conference contribution
posted on 2010-12-01, 00:00 authored by Shamsul HudaShamsul Huda, H Jelinek, B Ray, A Stranieri, John YearwoodJohn YearwoodCardiovascular autonomic neuropathy (CAN) is one of the important causes of mortality among diabetes patients. Statistics shows that more than 22% of people with type 2 diabetes mellitus suffer from CAN and which in turn leads to cardiovascular disease (heart attack, stroke). Therefore early detection of CAN could reduce the mortality. Traditional method for detection of CAN uses Ewing's algorithm where five noninvasive cardiovascular tests are used. Often for clinician, it is difficult to collect data from for the Ewing Battery patients due to onerous test conditions. In this paper, we propose a hybrid of wrapper-filter approach to find novel features from patients' ECG records and then generate decision rules for the new features for easier detection of CAN. In the proposed feature selection, a hybrid of filter (Maximum Relevance, MR) and wrapper (Artificial Neural Net Input Gain Measurement Approximation ANNIGMA) approaches (MR-ANNIGMA) would be used. The combined heuristics in the hybrid MRANNIGMA takes the advantages of the complementary properties of the both filter and wrapper heuristics and can find significant features. The selected features set are used to generate a new set of rules for detection of CAN. Experiments on real patient records shows that proposed method finds a smaller set of features for detection of CAN than traditional method which are clinically significant and could lead to an easier way to diagnose CAN. © 2010 IEEE.
History
Pagination
297-302Location
Brisbane, Qld.Publisher DOI
Start date
2010-12-07End date
2010-12-10ISBN-13
9781424471768Publication classification
EN.1 Other conference paperTitle of proceedings
Proceedings of the 2010 6th International Conference on Intelligent Sensors, Sensor Networks and Information Processing, ISSNIP 2010Publisher
IEEEPlace of publication
Piscataway, N.J.Usage metrics
Categories
No categories selectedKeywords
Licence
Exports
RefWorksRefWorks
BibTeXBibTeX
Ref. managerRef. manager
EndnoteEndnote
DataCiteDataCite
NLMNLM
DCDC