Heon Gyu Lee

A data mining approach for dyslipidemia disease prediction using carotid arterial feature vectors

In this paper, we proposed a useful methodology for the diagnosis of dyslipidemia disease by using novel various features of carotid arterial wall thickness. We measured and tested intima-media thickness of carotid arteries and used them as diagnostic feature vectors. In order to evaluate extracted various features, we tested on five classification methods and evaluated performance of classifiers. As a result, SVM and Neural Network algorithms (about 92%–98% goodness of fit) outperformed the other classifiers on those selected features.

Emerging Patterns Based Methodology for Prediction of Patients with Myocardial Ischemia

Heart disease is the one of the significant health problem in the world. Recently, most serious problem caused by it is that the patient becomes younger. Therefore, it is very important and necessary to find the early symptoms of heart problems for better treatment and effective methodology for predicting the disease. Data mining is the one of the efficient approaches. However, there are still some tasks have to be solved. One is that the result should make it easy to explain the relationship between class label and predictors for the heart disease data. In this paper, redefined T-tree algorithm is used to mine the emerging patterns to perform the work and solve the problem. Also, the aggregate score is considered to build classifier for the prediction work. The algorithms CMAR, CPAR, C4.5 and our method are applied to the dataset and the proposed method shows the better accuracy than others (The accuracy is between 75% to 85%)

Real-Time Diagnosis System Using Incremental Emerging Pattern Mining

Currently, as a effort to reduce a rate of death by cardiovascular diseases, a lot of researches have been studied regarding real-time diagnosis system. So, we implement a prototype which is contained of stream data processor and incremental data mining module for automatic diagnosis of cardiovascular diseases. In the prototype, (i)ECG signal data which is transported from body-attached sensor is collected and pre-processed, and (ii)diagnosis features of the bio-signal data are extracted. And the patients are automatically diagnosed using the incremental emerging pattern mining module, then (iii)the diagnosis result is provided for the doctor in charge of the patients via a web application in order to manage the medical history of each patient. So, the prototype is able to diagnose and predict patient state on real-time, automatically.

Spatio-temporal mining for power load forecasting in GIS-AMR load analysis model

A spatio-temporal mining technique is used to predict power load patterns for a voltage transformer. It is applied from load data measured every thirty minutes and a GIS-AMR database collected by a transformers load measurement system over a wireless network. The proposed approach in this paper consists of three stages, (i) data preprocessing: noise or outlier is removed and the continuous attribute-valued features are transformed to new features (feature extraction and discretization), (ii) cluster analysis: SOMs (Self Organizing Maps) clustering is used to label the class and (iii) classification: we used and evaluated classification rules using spatio-temporal mining to build a suitable load forecasting model. In order to evaluate the result of classification, derived class labels from clustering and other features are used as input to build classification rules including time and spatial factors. Lastly, the result of our experiments is presented.

Emerging Pattern Based Classification for Automated Non-safe Power Line Detection

In this paper, we proposed an automated load patterns classification technique to detect potentially non-safe power lines to prevent the power failure and to control the power distribution system efficiently. In operating power distribution, if there is overload in the fixed load capability of a power line, the power line will be breakdown, and that accident causes significant financial damages. For the prevention of power cut, we extracted the load shape feature according to the characteristic of electric consumption in Korea, and detected the anomalous patterns of non-safe power lines group using classifier based on EPs (emerging patterns) [1]. The discovered EPs have high support in non-safe power lines group and have low support in the normal group. In order to evaluate our classification method, power load data and information of 401 power lines are used during Feb. 2007, and compared with several existing classification methods. As a result of our experiments, the overall accuracy of EPs-based classification method about power load data was turned out to be about 91.75%. And the accuracy of non-safe power line group was about 96%.