• Korean Journal of Materials Research
• /
• v.34 no.7
• /
• pp.363-369
• /
• 2024

In this study, the magnetocaloric effect and transition temperature of bulk metallic glass, an amorphous material, were predicted through machine learning based on the composition features. From the Python module 'Matminer', 174 compositional features were obtained, and prediction performance was compared while reducing the composition features to prevent overfitting. After optimization using RandomForest, an ensemble model, changes in prediction performance were analyzed according to the number of compositional features. The R2 score was used as a performance metric in the regression prediction, and the best prediction performance was found using only 90 features predicting transition temperature, and 20 features predicting magnetocaloric effects. The most important feature when predicting magnetocaloric effects was the 'Fe' compositional ratio. The feature importance method provided by 'scikit-learn' was applied to sort compositional features. The feature importance method was found to be appropriate by comparing the prediction performance of the Fe-contained dataset with the full dataset.

• KIISE Transactions on Computing Practices
• /
• v.20 no.10
• /
• pp.549-554
• /
• 2014

Handwritten character recognition data is usually imbalanced because it is collected from the natural language sentences written by different writers. The imbalanced data can cause seriously negative effect on the performance of most of machine learning algorithms. But this problem is typically ignored in handwritten character recognition, because it is considered that most of difficulties in handwritten character recognition is caused by the high variance in data set and similar shapes between characters. We propose the oversampling-based ensemble learning methods to solve imbalanced data problem in handwritten character recognition and to improve the recognition accuracy. Also we show that proposed method achieved improvements in recognition accuracy of minor classes as well as overall recognition accuracy empirically.

• Journal of Korean Institute of Industrial Engineers
• /
• v.28 no.1
• /
• pp.112-127
• /
• 2002

A useful pattern is a pattern that contributes much to learning. For a classification problem those patterns near the class boundary surfaces carry more information to the classifier. For a regression problem the ones near the estimated surface carry more information. In both cases, the usefulness is defined only for those patterns either without error or with negligible error. Using only the useful patterns gives several benefits. First, computational complexity in memory and time for learning is decreased. Second, overfitting is avoided even when the learner is over-sized. Third, learning results in more stable learners. In this paper, we propose a pattern 'utility index' that measures the utility of an individual pattern. The utility index is based on the bias and variance of a pattern trained by a network ensemble. In classification, the pattern with a low bias and a high variance gets a high score. In regression, on the other hand, the one with a low bias and a low variance gets a high score. Based on the distribution of the utility index, the original training set is divided into a high-score group and a low-score group. Only the high-score group is then used for training. The proposed method is tested on synthetic and real-world benchmark datasets. The proposed approach gives a better or at least similar performance.

• The Journal of Korea Robotics Society
• /
• v.15 no.4
• /
• pp.385-391
• /
• 2020

As deep learning technology has been developed and applied to various fields, it is gradually changing from an existing single image based application to a video based application having a time base in order to recognize human behavior. However, unlike 2D CNN in a single image, 3D CNN in a video has a very high amount of computation and parameter increase due to the addition of a time axis, so improving accuracy in action recognition technology is more difficult than in a single image. To solve this problem, we investigate and analyze various techniques to improve performance in 3D CNN-based image recognition without additional training time and parameter increase. We propose a time base ensemble using the time axis that exists only in the videos and an ensemble in the input frame. We have achieved an accuracy improvement of up to 7.1% compared to the existing performance with a combination of techniques. It also revealed the trade-off relationship between computational and accuracy.

• Journal of Korea Multimedia Society
• /
• v.16 no.9
• /
• pp.1005-1017
• /
• 2013

Voice activity detection is very important process that voice activity separated form noisy speech signal for speech enhance. Over the past few years, many studies have been made on voice activity detection, but it has poor performance in low signal to noise ratio environment or fickle noise such as car noise. In this paper, it proposed new voice activity detection algorithm using ensemble variance based on wavelet band entropy and soft thresholding method. We conduct a survey in a lot of signal to noise ratio environment of car noise to evaluate performance of the proposed algorithm and confirmed performance of the proposed algorithm.

• Journal of Korean Society on Water Environment
• /
• v.28 no.3
• /
• pp.359-366
• /
• 2012

Accurate predictions about the water quality of a river have great importance in identifying in-stream flow and water supply requirements and solving relevant environmental problems. In this study, the effect of water release from upstream dam on the downstream water quality has been investigated by applying a hydological model combined with QUAL2E to Geum River basin. The ESP (Ensemble Stream Prediction) method, which has been validated and verified by lots of researchers, was used to predict reservoir and tributary inflow. The input parameters for a combined model to predict both hydrological characteristics and water quality were identified and optimized. In order to verify the model performance, the simulated result at Gongju station, located at the downstream from Daecheong Dam, has been compared with measured data in 2008. As a result, it was found that the proposed model simulates well the values of BOD, T-N, and T-P with an acceptable reliability.

• Journal of Biomedical Engineering Research
• /
• v.35 no.3
• /
• pp.42-49
• /
• 2014

We propose an individual identification method using a single-lead electrocardiogram signal. In this paper, lead I ECG is measured from subjects in various physical and psychological states. We performed a noise reduction for lead I signal as a preprocessing stage and this signal is used to acquire the representative beat waveform for individuals by utilizing the ensemble average. From the P-QRS-T waves, features are extracted to identify individuals, 19 using the duration and amplitude information, and 16 from the QRS complex acquired by applying Pan-Tompkins algorithm to the ensemble averaged waveform. To analyze the effect of each feature and to improve efficiency while maintaining the performance, Relief-F algorithm is used to select features from the 35 features extracted. Some or all of these 35 features were used in the support vector machine (SVM) learning and tests. The classification accuracy using the entire feature set was 98.34%. Experimental results show that it is possible to identify a person by features extracted from limb lead I signal only.

• Journal of Electrical Engineering and Technology
• /
• v.7 no.1
• /
• pp.115-123
• /
• 2012

A video analysis system used to detect events in video streams generally has several processes, including object detection, object trajectories analysis, and recognition of the trajectories by comparison with an a priori trained model. However, these processes do not work well in a complex environment that has many occlusions, mirror effects, and/or shadow effects. We propose a new approach to a context-aware video surveillance system to detect predefined contexts in video streams. The proposed system consists of two modules: a feature extractor and a context recognizer. The feature extractor calculates the moving energy that represents the amount of moving objects in a video stream and the stationary energy that represents the amount of still objects in a video stream. We represent situations and events as motion changes and stationary energy in video streams. The context recognizer determines whether predefined contexts are included in video streams using the extracted moving and stationary energies from a feature extractor. To train each context model and recognize predefined contexts in video streams, we propose and use a new ensemble classifier based on the AdaBoost algorithm, DAdaBoost, which is one of the most famous ensemble classifier algorithms. Our proposed approach is expected to be a robust method in more complex environments that have a mirror effect and/or a shadow effect.

• Journal of Korea Multimedia Society
• /
• v.19 no.8
• /
• pp.1530-1537
• /
• 2016

Recently, solar power generation shows the significant growth in the renewable energy field. Using the short-term prediction, it is possible to control the electric power demand and the power generation plan of the auxiliary device. However, a short-term prediction can be used when you know the weather forecast. If it is not possible to use the weather forecast information because of disconnection of network at the island and the mountains or for security reasons, the accuracy of prediction is not good. Therefore, in this paper, we proposed a system capable of short-term prediction of solar power generation amount by using only the weather information that has been collected by oneself. We used temperature, humidity and insolation as weather information. We have applied a moving average to each information because they had a characteristic of time series. It was composed of min, max and average of each information, differences of mutual information and gradient of it. An artificial neural network, SVM and RBF Network model was used for the prediction algorithm and they were combined by Ensemble method. The results of this suggest that using a moving average during pre-processing and ensemble prediction models will maximize prediction accuracy.

• Journal of the Korea Computer Graphics Society
• /
• v.28 no.1
• /
• pp.1-9
• /
• 2022

Non-small cell lung cancer (NSCLC) accounts for a high proportion of 85% among all lung cancer and has a significantly higher mortality rate (22.7%) compared to other cancers. Therefore, it is very important to predict the prognosis after surgery in patients with non-small cell lung cancer. In this study, the types of preoperative chest CT image patches for non-small cell lung cancer patients with tumor as a region of interest are diversified into five types according to tumor-related information, and performance of single classifier model, ensemble classifier model with soft-voting method, and ensemble classifier model using 3 input channels for combination of three different patches using pre-trained ResNet and EfficientNet CNN networks are analyzed through misclassification cases and Grad-CAM visualization. As a result of the experiment, the ResNet152 single model and the EfficientNet-b7 single model trained on the peritumoral patch showed accuracy of 87.93% and 81.03%, respectively. In addition, ResNet152 ensemble model using the image, peritumoral, and shape-focused intratumoral patches which were placed in each input channels showed stable performance with an accuracy of 87.93%. Also, EfficientNet-b7 ensemble classifier model with soft-voting method using the image and peritumoral patches showed accuracy of 84.48%.