Statlog Heart Research Articles

Feature selection based on the center of gravity of BSWFMs using NEWFM

Feature selection has commonly been used to remove irrelevant features and improve classification performance. Some of features are irrelevant to the learning process; therefore to remove these irrelevant features not only decreases training and testing times, but can also improve learning accuracy. This study proposes a novel supervised feature selection method based on the bounded sum of weighted fuzzy membership functions (BSWFM) and Euclidean distances between their centers of gravity for decreasing the computational load and improving accuracy by removing irrelevant features. This study compares the performance of a neural network with a weighted fuzzy membership function (NEWFM) without and with the proposed feature selection method. The superiority of the NEWFM with feature selection over NEWFM without feature selection was demonstrated using three experimental datasets from the UCI Machine Learning Repository: Statlog Heart, Parkinsons and Ionosphere. 13 features, 22 features, and 34 features were used as inputs for the NEWFM without feature selection and these resulted in performance accuracies of 85.6%, 86.2% and 91.2%, respectively, using Statlog Heart, Parkinsons and Ionosphere datasets. 10 minimum features, 4 minimum features and 25 minimum features were used as inputs for the NEWFM with feature selection and these resulted in performance accuracies of 87.4%, 88.2%, and 92.6%, respectively, using Statlog Heart, Parkinsons and Ionosphere datasets. The results show that NEWFM with feature selection performed better than NEWFM without feature selection.

Remote Heart Risk Monitoring System based on Efficient Neural Network and Evolutionary Algorithm

Objective: The objective of this paper is to predict the risk level of Heart Disease by applying Probabilistic Neural Network trained with Particle Swarm Optimization in case of Remote Health Monitoring. Methods: In order to achieve the aim of the activity, we propose hybrid model of Particle Swarm Optimization (PSO) and Probabilistic Neural Network (PNN). PSO is a population based meta-heuristic Evolutionary Algorithm (EA) whose goal is to explore the search space in order to find near – optimal solutions for feature selection. The optimal features selected can be used for prediction system to develop a classification model using probabilistic Neural Network. Results: First, we quantify the clinical data set from the UCI machine learning repository and measured the complexity. There are 13 attributes are used such as the age which identifies the age of the person, chest pain type has 4 values, serum cholesterol level, blood sugar, resting ECG results, serum cholesterol level, amount of heart rate achieved, x old peak, number of major vessels colored by fluoroscopy, slope of the peak exercise ST segment, thal, sex, height, weight and additional factor smoking. It has been shown that the time complexity of hybridizing PSO and PNN obtained the promising results compared to other two algorithms such as regression tree and PSO optimization. We also proposed the data mining process to deal with complexity, missing values and high dimensionality followed by incorporating the data mining functionalities like characterization, discrimination, association, classification, prediction and evolution analysis. The experiment carried out in Java on stat log heart disease data set performs better in all noise conditions. Conclusion: The performance was evaluated in terms of time complexity, accuracy, sensitivity and specificity and it proved that the hybrid model of PSO and PNN outperformed the Regression tree and PSO.

Knowledge mining from clinical datasets using rough sets and backpropagation neural network.

The availability of clinical datasets and knowledge mining methodologies encourages the researchers to pursue research in extracting knowledge from clinical datasets. Different data mining techniques have been used for mining rules, and mathematical models have been developed to assist the clinician in decision making. The objective of this research is to build a classifier that will predict the presence or absence of a disease by learning from the minimal set of attributes that has been extracted from the clinical dataset. In this work rough set indiscernibility relation method with backpropagation neural network (RS-BPNN) is used. This work has two stages. The first stage is handling of missing values to obtain a smooth data set and selection of appropriate attributes from the clinical dataset by indiscernibility relation method. The second stage is classification using backpropagation neural network on the selected reducts of the dataset. The classifier has been tested with hepatitis, Wisconsin breast cancer, and Statlog heart disease datasets obtained from the University of California at Irvine (UCI) machine learning repository. The accuracy obtained from the proposed method is 97.3%, 98.6%, and 90.4% for hepatitis, breast cancer, and heart disease, respectively. The proposed system provides an effective classification model for clinical datasets.

Analysis of Different Data Mining Tools using Classification, Clustering and Association Rule Mining

Now days in all fields to extract useful knowledge from data, data mining techniques like classification, clustering, association rule mining are useful. In data mining classification is categorization of different objects and Clustering is methodology using which we will be able to club objects of similar type. Another methodology like association rule mining (ARM) [1] is useful to find out association relationship among different objects. This paper compares performance of different data mining tools [2] like WEKA [3] , XLMiner [4] and KNIME [5] for these data mining techniques. We have used Statlog heart disease dataset [6] for analyzing performance of tools.

A new clustering algorithm based on Voronoi diagram

Clustering is a widely used data mining tool that has been paid enormous attention by the research community. We introduce here a new clustering algorithm which is based on Voronoi diagram. The biggest advantage of using Voronoi diagram is that it provides a natural means of partitioning space into sub-regions and therefore facilitates data clustering. In our approach, we exploit largest empty Voronoi circles to locate closer points represented by the Voronoi vertices referred as cluster prototypes. The points represented by these prototypes are then efficiently amalgamated by constructing new Voronoi diagram iteratively to produce the desired clusters. We perform excessive experiments on the synthetic as well as biological data, namely iris, wine, statlog heart, breast tissue, pima-Indians-diabetes, cloud, blood transfusion and yeast. The experimental results are compared with some standard clustering techniques. It is shown that the proposed algorithm performs better in discovering the complex clusters and able to detect the outliers.

Performance evolution of a newly developed general-use hybrid AIS-ANN system: AaA-response

In this study, we have developed a nonlinear recognition system in the artificial immune systems (AIS) field named `AaA-response (artificial neural network (ANN)-aided AIS-response)', which is different from previous AIS methods in that it uses a different modeling strategy in the formation of the memory response. Because it also uses ANNs in the determination of the correct output, it can be seen as a hybrid system that involves AIS and ANN approaches. Unlike the other AIS methods, AaA-response uses multiple system units (or antibodies) to form an output for a presented input. This property gives the proposed system the ability of producing the desired output values, other than just being a classification algorithm. That is, AaA-response can also be used as a regression method, like ANNs, by producing any output value for the given inputs. The parameter analyses of the system were conducted on an artificially generated dataset, the Chainlink dataset, and the important points in the parameter selection were emphasized. To investigate the performance of the system for real-world problems, the Iris dataset and Statlog Heart Disease dataset, taken from the University of California - Irvine machine learning repository, were used. The system, which obtained 99.33% classification accuracy on the Iris dataset, has shown an important performance superiority with regard to the classification accuracy to other methods in the literature by reaching 90.37% classification accuracy for the Statlog Heart Disease dataset.

An improved support vector machines model in medical data analysis

The support vector machine (SVM) technique is an emerging classification scheme that has been successfully employed in solving many classification problems. However, three main traits: features selection, dimension reduction and parameters selection, essentially influence the classification performance of SVM models. Therefore, this study developed an improved support vector machine (IMSVM) model using factor analysis (FA), kernel sliced inverse regression (KSIR) and honey-bee mating optimisation with genetic algorithms (HBMOG) to deal with feature selection, dimension reduction, and parameter selection issues, respectively, for SVM models. Then, the statlog heart dataset from the Center for Machine Learning and Intelligent Systems at the University of California, Irvine (UCI) was used to demonstrate the performance of the IMSVM model. Experimental results revealed that the IMSVM model can provide more accurate classification results than the results obtained by classification models in previous literature. Thus, the proposed model is a promising alternative for analysing medical data.

Use of Kernel Functions in Artificial Immune Systems for the Nonlinear Classification Problems

Due to the fact that there exist only a small number of complex systems in artificial immune systems (AISs) that solve nonlinear problems, there is a need to develop nonlinear AIS approaches that would be among the well-known solution methods. In this study, we developed a kernel-based AIS to compensate for this deficiency by providing a nonlinear structure via transformation of distance calculations in the clonal selection models of classical AIS to kernel space. Applications of the developed system were conducted on Statlog heart disease dataset, which was taken from the University of California, Irvine Machine-Learning Repository, and on Doppler sonograms to diagnose atherosclerosis disease. The system obtained a classification accuracy of 85.93% for the Statlog heart disease dataset, while it achieved a 99.09% classification success for the Doppler dataset. With these results, our system seems to be a potential solution method, and it may be considered as a suitable method for hard nonlinear classification problems.

Attribute weighting via genetic algorithms for attribute weighted artificial immune system (AWAIS) and its application to heart disease and liver disorders problems

An increasing number of algorithms and applications have coming into scene in the field of artificial immune systems (AIS) day by day. Whereas this increase is bringing successful studies, still, AIS is not an effective problem solver in some problem fields such as classification, regression, pattern recognition, etc. So far, many of the developed AIS algorithms have used a distance or similarity measure as the case in instance based learning (IBL) algorithms. The efficiency of IBL algorithms lies mainly in the weighting scheme they used. This weighting idea was taken as the objective of our study in that we used genetic algorithms to determine the weights of attributes and then used these weights in our previously developed Artificial Immune System (AWAIS). We evaluated the performance of new configuration (GA-AWAIS) on two medical datasets which were Statlog Heart Disease and BUPA Liver Disorders dataset. We also compared it with AWAIS for those problems. The obtained classification accuracy was very good with respect to both AWAIS and other common classifiers in literature.

Effect of feature-type in selecting distance measure for an artificial immune system as a pattern recognizer

In designing an artificial immune system (AIS) for a problem domain, one must select a distance measure to find the affinity between system units and input data after determining a representation type. Euclidean distance is a commonly used distance measure in many proposed methods and is selected intuitively or due to simplicity of implementation. But this selection must be done carefully by considering the properties of problem domain. For example, most problems use data vectors with discrete, real-valued and nominal feature values. Whereas Euclidean distance can be used in this kind of problems, some other similarity measures designed for these hybrid vectors would give better results. To call attention of AIS designer to this point, we have tested three distance criteria which are Euclidean distance, Manhattan distance, and hybrid similarity measure on a simple AIS for the classification of two medical dataset taken from the UCI machine learning repository. One of the datasets, Statlog heart disease, contains nominal, discrete and real-valued vectors while the other one, BUPA liver disorders dataset, consists of purely real-valued vectors. For Statlog dataset, the best classification result was obtained with hybrid similarity measure as expected because this dataset consists of three-types of features while results for BUPA dataset were not different so much for the used measures, which is also an expected result considering the nature of this dataset.