• Korean Journal of Radiology
• /
• v.22 no.3
• /
• pp.344-353
• /
• 2021

Objective: The mitotic count of gastrointestinal stromal tumors (GIST) is closely associated with the risk of planting and metastasis. The purpose of this study was to develop a predictive model for the mitotic index of local primary GIST, based on deep learning algorithm. Materials and Methods: Abdominal contrast-enhanced CT images of 148 pathologically confirmed GIST cases were retrospectively collected for the development of a deep learning classification algorithm. The areas of GIST masses on the CT images were retrospectively labelled by an experienced radiologist. The postoperative pathological mitotic count was considered as the gold standard (high mitotic count, > 5/50 high-power fields [HPFs]; low mitotic count, ≤ 5/50 HPFs). A binary classification model was trained on the basis of the VGG16 convolutional neural network, using the CT images with the training set (n = 108), validation set (n = 20), and the test set (n = 20). The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated at both, the image level and the patient level. The receiver operating characteristic curves were generated on the basis of the model prediction results and the area under curves (AUCs) were calculated. The risk categories of the tumors were predicted according to the Armed Forces Institute of Pathology criteria. Results: At the image level, the classification prediction results of the mitotic counts in the test cohort were as follows: sensitivity 85.7% (95% confidence interval [CI]: 0.834-0.877), specificity 67.5% (95% CI: 0.636-0.712), PPV 82.1% (95% CI: 0.797-0.843), NPV 73.0% (95% CI: 0.691-0.766), and AUC 0.771 (95% CI: 0.750-0.791). At the patient level, the classification prediction results in the test cohort were as follows: sensitivity 90.0% (95% CI: 0.541-0.995), specificity 70.0% (95% CI: 0.354-0.919), PPV 75.0% (95% CI: 0.428-0.933), NPV 87.5% (95% CI: 0.467-0.993), and AUC 0.800 (95% CI: 0.563-0.943). Conclusion: We developed and preliminarily verified the GIST mitotic count binary prediction model, based on the VGG convolutional neural network. The model displayed a good predictive performance.

• Journal of Intelligence and Information Systems
• /
• v.24 no.2
• /
• pp.111-124
• /
• 2018

This study aims to develop a classification model for predicting the occurrence of hyperlipidemia, one of the chronic diseases. Prior studies applying data mining techniques for predicting disease can be classified into a model design study for predicting cardiovascular disease and a study comparing disease prediction research results. In the case of foreign literatures, studies predicting cardiovascular disease were predominant in predicting disease using data mining techniques. Although domestic studies were not much different from those of foreign countries, studies focusing on hypertension and diabetes were mainly conducted. Since hypertension and diabetes as well as chronic diseases, hyperlipidemia, are also of high importance, this study selected hyperlipidemia as the disease to be analyzed. We also developed a model for predicting hyperlipidemia using SVM and meta learning algorithms, which are already known to have excellent predictive power. In order to achieve the purpose of this study, we used data set from Korea Health Panel 2012. The Korean Health Panel produces basic data on the level of health expenditure, health level and health behavior, and has conducted an annual survey since 2008. In this study, 1,088 patients with hyperlipidemia were randomly selected from the hospitalized, outpatient, emergency, and chronic disease data of the Korean Health Panel in 2012, and 1,088 nonpatients were also randomly extracted. A total of 2,176 people were selected for the study. Three methods were used to select input variables for predicting hyperlipidemia. First, stepwise method was performed using logistic regression. Among the 17 variables, the categorical variables(except for length of smoking) are expressed as dummy variables, which are assumed to be separate variables on the basis of the reference group, and these variables were analyzed. Six variables (age, BMI, education level, marital status, smoking status, gender) excluding income level and smoking period were selected based on significance level 0.1. Second, C4.5 as a decision tree algorithm is used. The significant input variables were age, smoking status, and education level. Finally, C4.5 as a decision tree algorithm is used. In SVM, the input variables selected by genetic algorithms consisted of 6 variables such as age, marital status, education level, economic activity, smoking period, and physical activity status, and the input variables selected by genetic algorithms in artificial neural network consist of 3 variables such as age, marital status, and education level. Based on the selected parameters, we compared SVM, meta learning algorithm and other prediction models for hyperlipidemia patients, and compared the classification performances using TP rate and precision. The main results of the analysis are as follows. First, the accuracy of the SVM was 88.4% and the accuracy of the artificial neural network was 86.7%. Second, the accuracy of classification models using the selected input variables through stepwise method was slightly higher than that of classification models using the whole variables. Third, the precision of artificial neural network was higher than that of SVM when only three variables as input variables were selected by decision trees. As a result of classification models based on the input variables selected through the genetic algorithm, classification accuracy of SVM was 88.5% and that of artificial neural network was 87.9%. Finally, this study indicated that stacking as the meta learning algorithm proposed in this study, has the best performance when it uses the predicted outputs of SVM and MLP as input variables of SVM, which is a meta classifier. The purpose of this study was to predict hyperlipidemia, one of the representative chronic diseases. To do this, we used SVM and meta-learning algorithms, which is known to have high accuracy. As a result, the accuracy of classification of hyperlipidemia in the stacking as a meta learner was higher than other meta-learning algorithms. However, the predictive performance of the meta-learning algorithm proposed in this study is the same as that of SVM with the best performance (88.6%) among the single models. The limitations of this study are as follows. First, various variable selection methods were tried, but most variables used in the study were categorical dummy variables. In the case with a large number of categorical variables, the results may be different if continuous variables are used because the model can be better suited to categorical variables such as decision trees than general models such as neural networks. Despite these limitations, this study has significance in predicting hyperlipidemia with hybrid models such as met learning algorithms which have not been studied previously. It can be said that the result of improving the model accuracy by applying various variable selection techniques is meaningful. In addition, it is expected that our proposed model will be effective for the prevention and management of hyperlipidemia.

• Proceedings of the KIEE Conference
• /
• 2004.11c
• /
• pp.496-498
• /
• 2004

An action-selection-mechanism is proposed to deal with sequential behaviors, where associations between some of stimulus and behaviors will be learned by a shortest-path-finding-based reinforcement team ins technique. To be specific, we define behavioral motivation as a primitive node for action selection, and then sequentially construct a network with behavioral motivations. The vertical path of the network represents a behavioral sequence. Here, such a tree fur our proposed ASM can be newly generated and/or updated. whenever a new sequential behaviors is learned. To show the validity of our proposed ASM, some experimental results on a "pushing-box-into-a-goal task" of a mobile robot will be illustrated.

• Journal of the Korean Operations Research and Management Science Society
• /
• v.30 no.4
• /
• pp.27-43
• /
• 2005

This paper presents a new hybrid data mining technique using error pattern modeling to improve classification accuracy when the data type of a target variable is binary. The proposed method increases prediction accuracy by combining two different supervised learning methods. That is, the algorithm extracts a subset of training cases that are predicted inconsistently by both methods, and models error patterns from the cases. Based on the error pattern model, the Predictions of two different methods are merged to generate final prediction. The proposed method has been tested using practical 10 data sets. The analysis results show that the performance of proposed method is superior to the existing methods such as artificial neural networks and decision tree induction.

• IEIE Transactions on Smart Processing and Computing
• /
• v.5 no.5
• /
• pp.358-365
• /
• 2016

We propose a novel application of random forest, a machine learning-based general classification algorithm, to analyze the influence of design attributes on the silicon-to-SPICE (S2S) gap. To improve modeling accuracy, we introduce magnification of learning data as well as randomization for the counting of design attributes to be used for each tree in the forest. From the automatically generated decision trees, we can extract the so-called importance and impact indices, which identify the most significant design attributes determining the S2S gap. We apply the proposed method to actual silicon data, and observe that the identified design attributes show a clear trend in the S2S gap. We finally unveil 10nm key fin-shaped field effect transistor (FinFET) structures that result in a large S2S gap using the measurement data from 10nm test vehicles specialized for model-hardware correlation.

• International Journal of Computer Science & Network Security
• /
• v.21 no.4
• /
• pp.272-276
• /
• 2021

Presently, Online / Offline Users are facing cyber attacks every day. These cyber attacks affect user's performance, resources and various daily activities. Due to this critical situation, attention must be given to prevent such users through cyber attacks. The objective of this research paper is to improve the IDS systems by using machine learning approach to develop a hybrid model which controls the cyber attacks. This Hybrid model uses the available KDD 1999 intrusion detection dataset. In first step, Hybrid Model performs feature optimization by reducing the unimportant features of the dataset through decision tree, support vector machine, genetic algorithm, particle swarm optimization and principal component analysis techniques. In second step, Hybrid Model will find out the minimum number of features to point out accurate detection of cyber attacks. This hybrid model was developed by using machine learning algorithms like PSO, GA and ELM, which trained the system with available data to perform the predictions. The Hybrid Model had an accuracy of 99.94%, which states that it may be highly useful to prevent the users from cyber attacks.

• The Bulletin of The Korean Astronomical Society
• /
• v.45 no.1
• /
• pp.50.2-50.2
• /
• 2020

The dark matter (DM) only simulations have been exploited to study e.g. the large scale structures and properties of a halo. In a baryon side, the high-resolution hydrodynamic simulation such as IllustrisTNG has helped extend the physics of gas along with stars and DM. However, the expansive computational cost of hydrodynamic simulations limits the size of a simulated universe whereas DM-only simulations can generate the universe of the cosmological horizon size approximately. I will introduce a pipeline to estimate baryonic properties of a galaxy inside a dark matter (DM) halo in DM-only simulations using a machine trained on high-resolution hydrodynamic simulations. An extremely randomized tree (ERT) algorithm is used together with multiple novel improvements such as a refined error function in machine training and two-stage learning. By applying our machine to the DM-only simulation of a large volume, I then validate the pipeline that rapidly generates a galaxy catalog from a DM halo catalog using the correlations the machine found in hydrodynamic simulations. I will discuss the benefits that machine-based approaches like this entail, as well as suggestions to raise the scientific potential of such approaches.

• Steel and Composite Structures
• /
• v.52 no.6
• /
• pp.695-711
• /
• 2024

Suspension bridges are critical to urban transportation, but those in earthquake-prone areas face unique challenges. In the event of a moderate or strong earthquake, conventional linear theory-based approaches for detecting bridge damage become inadequate. This study presents an efficient method for identifying damage in suspension bridges using time history nonlinear inelastic analysis. A practical advanced analysis program is employed to model cable-supported bridges with low computational cost, generating a dataset for four hybrid models: PSO-DT, PSO-RF, PSO-XGB, and PSO-CGB. These models combine decision tree (DT), random forest (RF), extreme gradient boosting (XGB), and categorical gradient boosting (CGB) with particle swarm optimization (PSO) to capture nonlinear correlations between displacement response and damage. Principal component analysis reduces dataset dimensions, and PSO selects the optimal model. A numerical case study of a suspension bridge under simulated earthquake conditions identifies PSO-XGB as the best model for predicting stiffness reduction. The results demonstrate the method's robustness for nonlinear damage detection in suspension bridges under earthquake excitation.

• Advances in materials Research
• /
• v.13 no.1
• /
• pp.63-86
• /
• 2024

Asphalt concrete (AC), is a mixture of bitumen and aggregates, which is very sensitive in the design of flexible pavement. In this study, the Marshall stability of the glass and carbon fiber bituminous concrete was predicted by using Artificial Neural Network (ANN), Support Vector Machine (SVM), Random Forest (RF), and M5P Tree machine learning algorithms. To predict the Marshall stability, nine inputs parameters i.e., Bitumen, Glass and Carbon fibers mixed in 100:0, 75:25, 50:50, 25:75, 0:100 percentage (designated as 100GF:0CF, 75GF:25CF, 50GF:50 CF, 25GF:75CF, 0GF:100CF), Bitumen grade (VG), Fiber length (FL), and Fiber diameter (FD) were utilized from the experimental and literary data. Seven statistical indices i.e., coefficient of correlation (CC), mean absolute error (MAE), root mean squared error (RMSE), relative absolute error (RAE), root relative squared error (RRSE), Scattering index (SI), and BIAS were applied to assess the effectiveness of the developed models. According to the performance evaluation results, Artificial neural network (ANN) was outperforming among other models with CC values as 0.9147 and 0.8648, MAE values as 1.3757 and 1.978, RMSE values as 1.843 and 2.6951, RAE values as 39.88 and 49.31, RRSE values as 40.62 and 50.50, SI values as 0.1379 and 0.2027 and BIAS value as -0.1 290 and -0.2357 in training and testing stage respectively. The Taylor diagram (testing stage) also confirmed that the ANN-based model outperforms the other models. Results of sensitivity analysis showed that the fiber length is the most influential in all nine input parameters whereas the fiber combination of 25GF:75CF was the most effective among all the fiber mixes in Marshall stability.

• KIPS Transactions on Software and Data Engineering
• /
• v.9 no.6
• /
• pp.187-194
• /
• 2020

This paper proposes a website fingerprinting method using ensemble learning over a Tor network that guarantees client anonymity and personal information. We construct a training problem for website fingerprinting from the traffic packets collected in the Tor network, and compare the performance of the website fingerprinting system using tree-based ensemble models. A training feature vector is prepared from the general information, burst, cell sequence length, and cell order that are extracted from the traffic sequence, and the features of each website are represented with a fixed length. For experimental evaluation, we define four learning problems (Wang14, BW, CW_T, CW_H) according to the use of website fingerprinting, and compare the performance with the support vector machine model using CUMUL feature vectors. In the experimental evaluation, the proposed statistical-based training feature representation is superior to the CUMUL feature representation except for the BW case.