مقایسه اجرای رگرسیون لجستیک، طبقه بندی و درخت رگرسیون و شبکه های عصبی برای پیش بینی بیماری عروق کرونر
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|24755||2008||9 صفحه PDF||سفارش دهید||5903 کلمه|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Expert Systems with Applications, Volume 34, Issue 1, January 2008, Pages 366–374
In this study, performances of classification techniques were compared in order to predict the presence of coronary artery disease (CAD). A retrospective analysis was performed in 1245 subjects (865 presence of CAD and 380 absence of CAD). We compared performances of logistic regression (LR), classification and regression tree (CART), multi-layer perceptron (MLP), radial basis function (RBF), and self-organizing feature maps (SOFM). Predictor variables were age, sex, family history of CAD, smoking status, diabetes mellitus, systemic hypertension, hypercholesterolemia, and body mass index (BMI). Performances of classification techniques were compared using ROC curve, Hierarchical Cluster Analysis (HCA), and Multidimensional Scaling (MDS). Areas under the ROC curves are 0.783, 0.753, 0.745, 0.721, and 0.675, respectively for MLP, LR, CART, RBF, and SOFM. MLP was found the best technique to predict presence of CAD in this data set, given its good classificatory performance. MLP, CART, LR, and RBF performed better than SOFM in predicting CAD in according to HCA and MDS.
Coronary artery disease (CAD) is a major worldwide health problem with its incidence and mortality rates (Backer et al., 2003). Many risk factors are known to play role in pathogenesis of CAD and myocardial infarction. Family history, smoking, hypertension, hypercholesterolemia, diabetes mellitus and obesity have been described as the major risk factors for CAD (Burke et al., 1997, Celermajer et al., 1993, Chobanian et al., 2003, Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults, 2001, Fuster, 1994, Haskell et al., 1994, Herrington et al., 2003, Higgins et al., 1992, Rahman et al., 2004, The Expert Committee on the Diagnosis and Classification of Diabetes Mellitus, 2003 and Zeiher et al., 1993). Identification of risk factors in CAD is essential for the management and follow-up of CAD. Numerous cross-sectional angiography studies have reported correlations of one or more major risk factors for myocardial infarction with the presence of CAD. Predicting the outcome of a disease is one of the most interesting and challenging tasks in which to develop data mining applications. Some applications of neural networks in cardiology were performed for the analysis of heart sounds, analysis of cardiac arrhythmias, the detection of ventricular ectopic activity, and the detection of atrial fibrillation. Furthermore, the development of implantable devices for treatment of life-threatening arrhythmias has simulated intracardiac rhythm classification using neural networks. Neural networks have been trained to recognize ST-T segment changes, to recognize CAD in general, to predict the number of vessels involved and to identify three-vessel and mainstem disease, even at rest (Dassen, Egmont-Petersen, & Mulleneers, 1998). Few works have been published on the comparison of classification techniques in different areas. Moisen and Frescino (2002) compared linear models, generalized additive models, classification and regression tree (CART), Multivariate Additive Regression Splines (MARS), and artificial neural networks for mapping forest characteristics in the Interior Western United States using forest inventory field data and ancillary satellite-based information. Ture, Kurt, Kurum, and Ozdamar (2005) compared various classification techniques to predict control and hypertension groups. They created models using logistic regression (LR), flexible discriminant analysis (FDA), FDA with MARS (FDA/MARS), chi-squared automatic interaction detector (CHAID), quick unbiased efficient statistical tree (QUEST), CART, radial basis function (RBF) and multi-layer perceptron (MLP) to predict control and hypertension groups. Delen, Walker, and Kadam (2004) compared LR, decision tree (C5) and artificial neural networks for predicting the survivability of diagnosed cases for breast cancer. Stark and Pfeiffer (1999) compared LR, classification tree algorithms (ID3, C4.5, CHAID, CART) and artificial neural networks to solve classification problems in complex data sets in veterinary epidemiology. Colombet et al. (2000) evaluated the implementation and performance of CART and artificial neural networks comparatively with a LR model, in order to predict the risk of cardiovascular disease in a real database. King, Feng, and Sutherland (1995) compared symbolic learning (CART, C4.5, NewID, AC2, ITrule, Cal5, and CN2), statistics (Naı¨ve Bayes, k-nearest neighbor, kernel density, linear discriminant, quadratic discriminant, LR, projection pursuit, and Bayesian networks), and neural networks (back-propagation and RBF) algorithms on twelve datasets with respect to large real-world problems. The purpose of this study is to compare performances of classification techniques in order to predict the presence of CAD. We have created models using LR, CART, neural networks algorithms (RBF, MLP, and self-organizing feature maps (SOFM)) that they are often used for classification problems. LR is useful for situations in which you want to be able to predict the presence or absence of a characteristic or outcome based on values of set of independent variables which are continuous, categorical, or both. Furthermore, it assumes that measures of dependent variables are independently and randomly sampled, all potentially relevant independent variables are in the model and all independent variables in the model are relevant (Hosmer and Lemeshow, 2000 and Kleinbaum, 1994). CART is inherently non-parametric that no assumptions are made regarding the underlying distribution of values of the predictor variables. Thus, CART can handle numerical data that are highly skewed or multi-modal, as well as categorical predictors with either ordinal or non-ordinal structure (Breiman, Friedman, Olshen, & Stone, 1984). Neural networks have been used to model medical and functional outcomes of dangerous disease. They have become a popular tool for classification, as they are very flexible, not assuming any parametric form for distinguishing between categories (Lee, 2001). Performances of classification techniques were compared using ROC curve, Hierarchical Cluster Analysis, and Multidimensional Scaling.