Academic
Publications
Exploring novel features and decision rules to identify cardiovascular autonomic neuropathy using a hybrid of wrapper-filter based feature selection

Exploring novel features and decision rules to identify cardiovascular autonomic neuropathy using a hybrid of wrapper-filter based feature selection,1

Exploring novel features and decision rules to identify cardiovascular autonomic neuropathy using a hybrid of wrapper-filter based feature selection  
BibTex | RIS | RefWorks Download
Cardiovascular 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 MR-ANNIGMA 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.
Cumulative Annual
View Publication
The following links allow you to view full publications. These links are maintained by other sources not affiliated with Microsoft Academic Search.