Earticle

This study investigates energy efficiency issues of device-to-device (D2D) communications in heterogeneous networks. To minimize the total transmitted power, an approach based on Q-learning together with adaptive ? -greedy is proposed to optimize the connection of user equipment (UE) with base station (BS) or access point (AP). The proposed adaptive ? -greedy can conduct the adequate exploration and exploitation operations for effective optimization. Simulation results indicate that in the single-cell scenario, the proposed method can attain performance close to the best solution.

Purpose: The purpose of this study was to identify the effects simulation-based education combined team-based learning (SBE combined TBL) compared to simulation-based education (SBE) on undergraduate nursing students. Methods: A non-equivalent control group design with pre-and posttest measures was used. The participants in the study were 181 students. The SBE combined TBL group consisted of 84 senior students in 2013, and the SBE group consisted of 97 seniors in 2014. Collected data were analyzed using chi-square, independent t-test and ANCOVA with the statistical package SPSS 22.0 for Windows. Results: There was a significant improvement in communication skills, nursing performance confidence, team efficacy, and team performance scores in the SBE combined TBL group compared to the SBE group (t=2.45, p=.015; F=4.30, p=.040; t=3.06, p=.003; t=8.77, p<.001). However, there was no statistically significant difference in self-directed learning between the groups. Conclusion: SBE combined TBL compared to SBE is an effective teaching and learning method to enhance various positive educational outcomes for nursing students. Therefore, we suggest that future studies investigate the development of an integrated course in which team-based learning is applied to theoretical sessions and simulation-based training.

Background Logistics service workers face elevated injury rates, with ankle-related incidents comprising 23% of workplace injuries. Traditional binary foot posture classifications may oversimplify the complex biomechanical relationships affecting dynamic balance performance in occupational settings. Purpose To apply unsupervised machine learning clustering to identify distinct foot posture phenotypes among logistics service workers and compare dynamic balance performance between identified clusters using the Y-Balance Test. Study design Cross-sectional observational study Methods A total of 112 logistics service workers were analyzed using K-means clustering based on age, body mass index, work duration, navicular drop, and resting calcaneal stance position. Silhouette score analysis determined optimal cluster number. Dynamic balance was assessed using the Y-Balance Test, measuring reach distances in anterior, posteromedial, and posterolateral directions. Results Four distinct phenotypes emerged: rearfoot valgus-dominant pronated (n=22), midfoot collapse-dominant pronated (n=32), age-related (n=23), and supinated (n=35) foot types. Significant differences in dynamic balance performance were observed in the posterolateral direction (F=3.900, p=0.011). The supinated phenotype demonstrated superior posterolateral reach performance (77.76±13.63%) compared to midfoot collapse-dominant (66.18±16.37%, p=0.003) and age-related phenotypes (67.78±14.97%, p=0.018). Conclusions Unsupervised machine learning successfully identified naturally occurring foot posture phenotypes with distinct dynamic balance characteristics. Midfoot collapse-dominant pronation demonstrated greater balance impairments than rearfoot valgus patterns, supporting the implementation of phenotype-specific interventions for workplace injury prevention in logistics workers.

Reinforcement learning is an area of machine learning that studies how an intelligent agent takes actions in a given environment to maximize the cumulative reward. In this paper, we propose a new MAC protocol based on the Q-learning technique of reinforcement learning to improve the performance of the IEEE 802.11 wireless LAN CSMA/CA MAC protocol. Furthermore, the operation of each access point (AP) and station is proposed. The AP adjusts the value of the contention window (CW), which is the range for determining the backoff number of the station, according to the wireless traffic load. The station improves the performance by selecting an optimal backoff number with the lowest packet collision rate and the highest transmission success rate through Q-learning within the CW value transmitted from the AP. The result of the performance evaluation through computer simulations showed that the proposed scheme has a higher throughput than that of the existing CSMA/CA scheme.

This paper discusses the performance of linear regression, regression tree, support vector regression, and ensemble learners in modelling airflow between two spaces based on accuracy and training time. To obtain training data, different scenarios from an existing computational fluid dynamics (CFD) model are simulated via transient analysis using Cradle scSTREAM. The raw dataset is transformed to having time step sizes of 2.5s, 5.0s, and 50.0s. Feature scaling is also employed on the each data set using both min-max scaling and z-score methods for a total of 9 datasets. Hyperparameters according to machine learning (ML) algorithms are varied such that 15 ML models across the four algorithms are developed. The results show that the regression trees perform the best over all other algorithms, with all models maintaining R2 values above 0.95 at the different datasets. On the other hand, as expected, all linear models demonstrated poor performance compared to nonlinear models. Data resolution affects model accuracy and training time, with accuracy declining slightly as time step size increased. It is also found that there is no significant effect of feature scaling. Lastly, ML models yield substantially cheaper simulation costs than CFD to simulate airflow.

This research delved into the impact of SRL strategies-based instruction on EFL students’ L2 writing performance, academic self-efficacy, and use of SRL strategies. To this end, 50 Iranian EFL learners, including 25 males and 25 females, attended the study. The experimental group was provided with a four-month SRL strategies-based intervention which followed the Self-Regulated Strategy Development model adapted from Harris and Graham (1996), and the control group took the traditional writing course needed by the university curriculum with the same length. All the students were required to accomplish pre- and post-writing tests as well as self-report questionnaires, including the Writing Strategies for Self-Regulated Learning Questionnaire and the second Language Writer Self-Efficacy Scale, at the beginning and the end of the instruction. The outcomes showed that the experimental group outperformed the control group in the post-writing test. Experimental group 's levels of linguistic self-efficacy and performance self-efficacy were remarkably higher than their counterparts in the control group. The study further implies that the instruction might contribute to the increased application of SRL strategies in EFL writing classes in a way that students might become more self-directed and self-regulated in their overall language performance in general and their written performances in particular.

و یلیصحت یدمآراکدوخ ،یاھ یژتارتسا زا هدافتسا تسا ھتخادرپ .زومآ نابز هاجنپدندوب مق هاگشناد رد یسیلگنا تایبدا ھتشر رد لیصحت لاح رد ھک نیا رددندرک تکرش قیقحتشزومآ تحت هام راھچ یشیامزآ هورگ . یمیظنتدوخ دربھار ھعسوت لدم قبط ھک تفرگ رارق روحم یژتارتسا ) ماھارگ و سیرھ1996 تدم نامھ ھب لرتنک هورگ یارب هاگشناد یسرد ھمانرب قبط یتنس یراتشون هرود و دوب ( راھچ ھئارا هامشیپ ات دش ھتساوخ نایوجشناد ھمھ زا .دشسپ و نومزآنومزآشسرپ و راتشون یاھھماندوخ یاھ شرازگ شسرپ لماش یھد ھمانیمیظنتدوخ یریگدای یارب راتشون یاھ یژتارتسا هورگ ھک داد ناشن جیاتن .دنھد ماجنا ار مود نابز هدنسیون یدمآراکدوخ رایعم و سپ رد لرتنک هورگ زا یرتھب درکلمع یشیامزآ هورگ یدرکلمع یدمآراکدوخ و ینابز یدمآراکدوخ حطس .تشاد راتشون نومزآشیامزآ یژتارتسا زا هدافتسا شیازفا ثعاب روحم یژتارتسا شزومآ زا هدافتسا ھک دش شرازگ نینچمھ .تشاد یریگمشچ شیازفا ی یاھلاعف و یم یشزومآ نینچ ھک تسا نیا زا یکاح ام جیاتن .دش روکذم یاھ یژتارتسا زا هدافتسا رد وجشناد ندش رت ھب دناوت نابز نازومآشیب یاکتا ردیگدای ھب رتو هدرک کمک ملعم زا یریگدای یاجب یدرف یرزومآ نابز یم ھنلااعف ار دوخ یریگدای دنیآرف دناوت رت .دنک یبایزرا

This study empirically verified the factors affecting the learning outcomes of English as a Foreign Language (EFL) learners in a smart learning (s-learning) environment. Based on social cognitive theory (SCT), six independent constructs (facilitating conditions, social influence, positive attitude toward s-learning, self-efficacy, social interaction, and knowledge) were proposed in the research model. The data collected from 455 s-learning EFL learners were analyzed in order to test the causal relationships between these six constructs and s-learning performance. For the analysis, the structural equation model (SEM) with AMOS 22.0 was used, and all six constructs were found to have a significant effect on s-learning performance. Important implications are discussed based on the study results.

조류는 생태계를 구성하는 중요한 요소이다. 그러나 남조류의 과도한 성장은 하천환경에 다양한 악영향을 발생시키고 규조류는 상수원과 정수장 공정관리에 영향을 미친다. 지속적이고 효율적인 조류 관리를 위해 조류 모니터링이 중요 하다. 본 연구에서는 You Only Look Once (YOLO)의 최신 알고리즘 YOLO v8을 사용하여 조류경보제 기준에 사 용하는 유해 남조류 4종과 정수처리공정에 영향이 큰 규조류 1종 총 5종의 이미지를 분류하는 이미지 분류모형을 구 축하였다. 기본모형의 mAP는 64.4로 분석되었다. 모형의 학습에 사용된 원본 이미지에 회전, 확대, 축소를 수행하여 이미지의 다양성을 높인 5가지 모형을 구축하여 입력자료로 사용된 이미지의 구성에 따른 모형 성능의 변화를 비교 하였다. 분석결과 회전, 확대, 축소를 모두 적용한 모형이 mAP 86.5로 가장 좋은 성능을 보이는 것을 확인하였다. 이 미지의 회전만을 적용한 모형, 회전과 확대를 적용한 모형, 이미지의 회전과 축소만를 적용한 모형의 mAP는 각각 85.3, 82.3, 83.8로 분석되었다.

Algae are an important component of the ecosystem. However, the excessive growth of cyanobacteria has various harmful effects on river environments, and diatoms affect the management of water supply processes. Algal monitoring is essential for sustainable and efficient algae management. In this study, an object detection model was developed that detects and classifies images of four types of harmful cyanobacteria used for the criteria of the algae alert system, and one diatom, Synedra sp.. You Only Look Once(YOLO) v8, the latest version of the YOLO model, was used for the development of the model. The mean average precision (mAP) of the base model was analyzed as 64.4. Five models were created to increase the diversity of the input images used for model training by performing rotation, magnification, and reduction of original images. Changes in model performance were compared according to the composition of the input images. As a result of the analysis, the model that applied rotation, magnification, and reduction showed the best performance with mAP 86.5. The mAP of the model that only used image rotation, combined rotation and magnification, and combined image rotation and reduction were analyzed as 85.3, 82.3, and 83.8, respectively.

본 연구는 데이터의 품질이 인공지능(AI) 성능에 미치는 영향을 검토한다. 이를 위해, 데이터 특성변수(Feature)의 유사도와 클래스(Class) 구성의 불균형을 고려한 모의실험(Simulation)을 통해 라벨링 오류 수준이 인공지능의 성능에 미치 는 영향을 비교 분석하였다. 그 결과, 특성변수 간 유사성이 높은 데이터에서는 특성 변수 간 유사성이 낮은 데이터에 비해 라 벨링 정확도에 더 민감하게 반응하였으며, 클래스 불균형이 증가함에 따라 인공지능 정확도가 급격히 감소되는 경향을 관찰 하였다. 이는 인공지능 학습데이터의 품질평가 기준 및 관련 연구를 위한 기초자료가 될 것이다.

The study investigates the impact of data quality on the performance of artificial intelligence (AI). To this end, the impact of labeling error levels on the performance of artificial intelligence was compared and analyzed through simulation, taking into account the similarity of data features and the imbalance of class composition. As a result, data with high similarity between characteristic variables were found to be more sensitive to labeling accuracy than data with low similarity between characteristic variables. It was observed that artificial intelligence accuracy tended to decrease rapidly as class imbalance increased. This will serve as the fundamental data for evaluating the quality criteria and conducting related research on artificial intelligence learning data.

뇌 구조 자기공명영상 데이터로부터 알츠하이머 환자인지 아닌지 분류하는 작업은 알츠하이머 질환의 조기 발견을 위해 도움이 되는 진단 기법이고, 최근 들어 딥러닝을 이용하여 진단의 정확도를 높이는 연구가 많이 진행되고 있 다. 3차원 영상으로부터 일부 슬라이스를 추출한 후 예측 모델 학습 및 검증을 진행하는 연구가 보고되고 있다. 본 논문에서는 슬라이스 선택 방식에 따른 차이가 뇌영상 알츠하이머 질환 딥러닝 분류의 성능에 어떤 영향을 주는지 비교 분석한다. Coronal 방향의 영상에서 세가지 방식을 비교한다. 엔트로피 순으로 선택하되 슬라이스 덩어리 별 로 선택하지 않는 방식과 덩어리 별로 선택하는 방식, 그리고 그룹간 L2-norm 값이 큰 순으로 선택하되 덩어리 단 위로 선택하는 방식을 비교한다. 5-fold 교차검증을 이용하여 모델 학습 및 검증을 수행한다. 딥러닝 모델로는 이미 지넷 데이터로 사전 학습한 가중치를 갖는 4가지 딥러닝 모델과 사전 학습하지 않은 딥러닝 모델을 이용하여 학습한 다. 검증 결과 엔트로피 기반 방식이 L2-norm 차이 기반의 방식에 비해 구현이 간단하면서도, 비슷한 예측 성능을 보여주고 있다.

The classification of Alzheimer patients vs. normals based on brain structural MRI data is helpful for early diagnosis of Alzheimer’s disease. Recently, numerous research studies investigated deep learning to improve diagnostic accuracy. Several research groups proposed methods for selecting slices from the 3D MRI data and utilized the selected slices as input to model training and validation. This paper addresses the effect of slice selection methods on deep learning classifiers’ performances in predicting Alzheimer’s disease. From the coronal slice orientation, we compared three slice selection methods: Entropy-based method without chunk selection and with chunk selection, and L2-norm-based method with chuck selection. Model development and validation were performed with a five-fold cross validation. We considered four deep convolutional neural network (CNN) models pretrained with ImageNet database as well as one deep CNN model without pretraining. Our results indicate that the entropy-based slice selection method is simpler than the L2-norm-based method in its implementation, and it produces performance similar to the L2-norm-based method.

뜰개는 해양에서 해수의 특성 및 흐름을 관측하기 위한 장비로서, 해수의 흐름 관측을 이용해 유출유 확산 예측을 위해 사용될 수 있다. 본 논문에서는 관측기관에서 사용하는 뜰개가 특정 시간 간격으로 관측한 바람 및 해수의 특성과 이동경로를 기계학습 기법들을 이용하여 학습시키고 예측하는 모델을 제안한다. 서포트벡터 회귀, 방사기저함수 네트워크, 가우시안 프로세스, 다층 퍼셉트론, 순환신경망을 이용하여 뜰개의 이동경로 예측 방법을 제시한다. 기존 MOHID 수치모델과 비교하여 각 기법별로 4 개의 사례중 3 개에서 성능이 개선되었으며, 가장 좋은 개선율을 보인 기법은 LSTM으로 평균 47.59% 개선되었다. 추후 연구에서는 배깅과 부스팅을 이용하여 가중치를 부여하여 정확도를 개선할 예정이다.

Drifter is an equipment for observing the characteristics of seawater in the ocean, and it can be used to predict effluent oil diffusion and to observe ocean currents. In this paper, we design models or the prediction of drifter trajectory using machine learning. We propose methods for estimating the trajectory of drifter using support vector regression, radial basis function network, Gaussian process, multilayer perceptron, and recurrent neural network. When the propose mothods were compared with the existing MOHID numerical model, performance was improve on three of the four cases. In particular, LSTM, the best performed method, showed the imporvement by 47.59% Future work will improve the accuracy by weighting using bagging and boosting.

Helicopter accidents due to spatial disorientation in low visibility conditions continue to persist as a major issue. These incidents often stem from human error, typically induced by stress, and frequently result in fatal outcomes. This study employs machine learning to analyze flight data and evaluate the efficacy of a flight illusion detection algorithm, laying groundwork for further research. This study collected flight data from approximately 20 pilots using a simulated flight training device to construct a range of flight scenarios. These scenarios included three stages of flight: ascending, level, and descent, and were further categorized into good visibility conditions and 0-mile visibility conditions. The aim was to investigate the occurrence of flight illusions under these conditions. From the extracted data, we obtained a total of 54,000 time-series data points, sampled five times per second. These were then analyzed using a machine learning approach.

인터넷의 발달과 비대면 서비스의 증가로 사용자들 간에 문자나 SNS로 소통하는 경우가 늘어나고 있다. 사용자에 의한 대량의 데이터가 발생하면서 사용자 정보나 의견을 분석하여 감정을 인식하는 연구가 활발하게 진행되고 있다. 그중에서도 텍스트 감정인식은 대부분 단어나 문장의 단일 감정을 인식하고 있다. 그러나 하나의 문장에도 여러 감 정이 복합적으로 존재하기 때문에 다중 감정인식 방법이 필요하다. 따라서 본 논문에서는 더욱 정확한 텍스트 감정 인식을 위해 데이터를 보정하는 방법과 딥러닝 기반의 다중 감정인식 방법을 적용한 감정인식 성능향상 방법을 제안 한다. 제안한 모델의 유용성을 확인하기 위해 딥러닝 모델을 비교 실험한 결과 Attention 모델을 사용했을 때 Accuracy가 76.7%로 가장 좋은 성능을 보였다.

With the development of the Internet and the increase of non-face-to-face services, the number of users communicating through text messages or SNS is increasing. As a large amount of data is generated by users, research on recognizing emotions by analyzing user information or opinions is being actively conducted. Among them, most of the text emotion recognition recognizes a single emotion of a word or sentence. However, since multiple emotions exist complexly in a single sentence, a multi-emotion recognition method is required. Therefore, in this paper, we propose a data correction method for more accurate text emotion recognition and a method to improve emotion recognition performance by applying a deep learning-based multi-emotion recognition method. As a result of comparing deep learning models to confirm the usefulness of the proposed model, when the attention model was used, Accuracy showed the best performance with 76.7%.

최근 들어 도로 노면 파손의 위치 정보를 수집하기 위한 영상 처리 기술에 대한 연구가 활 발히 진행되고 있다. 대표적으로 차량에 탑재가 가능한 스마트폰이나 블랙박스를 통해 영상을 얻고 이를 영상처리 알고리즘을 사용하여 인식하는 기술이 주로 사용된다. GPS 모듈과 연계하 여 실제 파손 위치를 파악할 때 가장 중요한 기술은 영상 처리 알고리즘인데, 근래에는 대부분 인공지능을 통한 알고리즘이 연구 주제로 주목받고 있다. 이와 같은 맥락에서 본 연구에서도 영역 기반의 합성곱 방식 계열의 객체인식 (Object Detection) 방법을 사용한 인공지능 영상 처 리 알고리즘에 대하여 논의하고자 한다. 도로 노면 파손 객체 인식 성능을 향상시키기 위하여 도로 노면 파손 영상 600여 장과 일반적인 도로 주행 영상 1500여 장으로 학습 데이터베이스를 구성하였다. 또한 배경 객체 인식 방법을 적용한 지도 학습을 수행하여 도로 노면 파손의 오탐 을 감소시켰다. 그 결과 동일한 테스트용 데이터베이스를 통해 알고리즘의 인식 성능을 mAP 평균값 기준 9.44%만큼 향상시킨 새로운 방법을 소개하고자 한다.

In recent years, image processing techniques for detecting road surface damaged spot have been actively researched. Especially, it is mainly used to acquire images through a smart phone or a black box that can be mounted in a vehicle and recognize the road surface damaged region in the image using several algorithms. In addition, in conjunction with the GPS module, the exact damaged location can be obtained. The most important technology is image processing algorithm. Recently, algorithms based on artificial intelligence have been attracting attention as research topics. In this paper, we will also discuss artificial intelligence image processing algorithms. Among them, an object detection method based on an region-based convolution neural networks method is used. To improve the recognition performance of road surface damage objects, 600 road surface damaged images and 1500 general road driving images are added to the learning database. Also, supervised learning using background object recognition method is performed to reduce false alarm and missing rate in road surface damage detection. As a result, we introduce a new method that improves the recognition performance of the algorithm to 8.66% based on average value of mAP through the same test database.

수요예측은 제품에 대한 수요량을 예측해 자원을 관리하기 위한 방법으로, 기업의 노동력과 예산 관리에 영향을 미 친다. 이러한 이유로 수요예측 모델의 성능 향상을 위한 연구가 주목을 받고 있다. 본 연구에서는 수요예측 성능 향 상을 위해 품목의 수요 패턴을 분석해 4가지 유형으로 구분하고, 각 유형에 적합한 모델을 제안한다. 성능 비교를 위해 사용한 데이터는 대한민국 공군 T-50 단일 기종의 수리 부속 품목의 분기 별 수요 데이터이다. 품목의 수요 패턴은 수요발생구간(average demand interval, ADI)과 변동 계수(coefficient of variation, CV)를 사용해 네 가지 smooth, lumpy, intermittent, erratic으로 구분하며 다양한 알고리즘으로 구현한 수요예측 모델의 성 능을 비교하기 위해 5가지 기계학습 알고리즘과 2 가지 딥러닝 알고리즘을 사용해 수요예측 모델을 구현한다. 기계 학습 알고리즘 중에는 앙상블 알고리즘인 random forest regression, adaboost, extra trees regression, bagging, gradient boosting regression 과 딥러닝 알고리즘인 long-short term memory(LSTM), deep neural network(DNN)을 사용한다. 수요 패턴에 따른 네 가지 유형에 적합한 모델을 선정해 수요예측 결과를 도 출한 경우가 일관된 모델을 사용한 경우에 비해 품목 정확도가 0.61%, 수량 정확도가 0.09 우수한 것을 확인할 수 있다. 제안하는 모델을 적용한다면 전문가의 효율적인 수요 관리가 이루어질 수 있을 것으로 기대한다.

Demand forecasting is a way to manage resources by forecasting demands for products, so it has direct impacts on corporate resources and budget management. Based on these reasons, research on improving forecasting performances of demand forecasting models. In this research, 4 demand patterns for items were analyzed to improve demand prediction performance, and the optimal model was proposed. The data used to compare the performance were the demand data from each quarter for maintenance items for a T-50 aircraft of Republic of Korea air force. First, the demand patterns for the items adopted average demand interval(ADI) and coefficient of variation(CV) and were categorized into smooth, lumpy, intermittent, and erratic items. In this research, to compare the performance of demand forecasting models derived from different algorithms, 5 types of machine learning algorithms and 2 types of deep learning algorithms were used to construct demand forecasting models. In machine learning algorithms, there are ensemble learning such as random forest regression, adaboost, extra trees regression, bagging, gradient boosting regression and deep learning algorithm such as long-short term memory(LSTM) and deep neural network(DNN). We can confirm that item accuracy is 0.61% and quantity accuracy is 0.09% better than that of consistent models when the demand forecast results are derived by selecting models suitable for four types according to demand patterns. We expect that efficient demand management by experts will be achieved if the application of the proposed model.

심전도 데이터의 자동 분석은 다양한 심장 질환을 진단하는데 매우 중요하고 임상의사의 진단 업무 효율성을 높이는 데 도움이 된다. 최근에 적은 수의 채널로 휴대가 가능한 웨어러블 심전도 기기가 개발되고, 이를 인공지능을 이용 하여 진단하는 기법들이 발표되고 있는 추세이지만, 리드 수 및 전처리 필터 타입이 클래스별로 심전도 분류에 미치 는 영향에 대한 연구는 부족하다. 본 논문에서는 PhysioNet 2021 Challenge에서 제공하는 12-리드 심전도 데이 터로 1D ResNet합성곱신경망모델을 이용하여 다중라벨로 분류하는 성능을 평가하였다.12-리드, 6-리드, 4-리드, 3-리드, 2-리드에 대해 리드별, 그리고 전처리를 사용하지 않는 방식, Pan-Tompkins(P-T)전처리 방식, 이산 웨 이브릿 변환 방식의 예측 성능을 비교하였다. 분산 분석을 적용한 결과, 클래스별 F1-score는 리드 수 및 전처리 방식에 대해 유의미한 차이를 보이지 않았다(p＞0.05). 클래스별로 다른 문턱값(threshold)을 적용하는 경우 일 정한 문턱값을 적용할 때보다 macro F1-score 값이 2-리드, P-T 전처리 기법의 경우 0.4448에서 0.4644로 증 가하였다. 본 다중라벨 심전도 데이터를 활용한 비교 분석 연구는 재현성을 도모하기 위해 소스 코드로 제공한다 (https://github.com/shjee20/1D-ecg).

Automatic analysis of electrocardiogram(ECG) data is crucial for the diagnosis of a variety of cardiovascular diseases, and it can help improve the efficiency of medical doctors’ tasks. Recently, portable and wearable ECG devices with a few of channels have been developed, and artificial intelligence-based techniques for disease diagnosis have been proposed in the literature. However, there is lack of studies investigating the effects of choosing lead counts and preprocessing methods on class-wise prediction performance. In this study, we utilized 12-lead ECG data provided by the PhysioNet 2021 Challenge and evaluated the performance of multi-label classifiers based on 1D ResNet convolutional neural network(CNN) models. We compared the prediction performance with a variety of lead choices(i.e., 12-lead, 6-lead, 4-lead, 3-lead, 2-lead) and preprocessing methods including no preprocessing, Pan-Tompkins filtering, and discrete wavelet transform.The analysis of variance(ANOVA) indicated that differences in class-wise F1-scores for the choice of lead counts and preprocessing methods were not statistically significant(p＞0.05). Compared to a fixed threshold setting, a class-dependent threshold setting improved macro F1-score from 0.448 to 0.4644 for the choice of the 2-lead and Pan-Tompkins preprocessing method. In order to promote reproducibility for comparative studies using multi-label ECG data, we provide source code at https://github.com/shjee20/1D-ecg

본 연구는 Zentner 음악 감성 모델을 기반으로 음악 감성 분류 성능을 향상시키는 머신러닝 기법을 탐구한다. GTZAN 데이터셋의 각 장르별 오디오에 Zentner모델의 음악 감성 레이블을 부여하고, 분류 수행을 위해 MFCC, ZCR, Chroma, Spectral Centroid, Harmony 등의 오디오 특징을 추출하였다. 분류 성능 비교를 위해 KNN, Decision Trees, Random Forest, XGBoost, LightGBM, SVM의 머신러닝 모델을 사용하여 음악 감성 분류 성능을 비교하였고. 추가적으로 재귀적 특성 제거(Recursive Feature Elimination)와 스태킹(Stacking) 기법을 사용하여 성능 향상 여부를 실험하였다. 그 결과, 개별 머신러닝 모델 분류에서는 LightGBM 모델에서 가장 높은 분류 성능을 보였으며, 스태킹 기법을 적용하는 경우에 개별 머신러닝 모델 분류 성능 평균 대비 14% 성능 향상을 확인하여 음악 감성 분류의 정밀도를 높이는 데 효과적임을 입증하였다.

This study explores music emotion classification using machine learning. The emotional categories are based on Zentner's music emotion model, and labels from this model were assigned to audio data from each genre in the GTZAN dataset. To perform classification, audio features such as MFCC, ZCR, Chroma, Spectral Centroid, and Harmony were extracted, and machine learning models including KNN, Decision Trees, Random Forest, XGBoost, LightGBM, and SVM were used to compare classification performance. In addition, the effectiveness of Recursive Feature Elimination and stacking techniques was tested, and improvements in classification performance were observed through stacking. The experimental results showed that among individual models, LightGBM achieved the highest classification performance. Moreover, applying the stacking technique led to a 14% improvement in performance on average compared to individual models.

본 논문에서는 얼굴 랜드마크를 사용하는 기존의 표정을 분류하는 방법이 아닌 이미지에 가장자리 검출 필터를 적용하여 얼굴 특징을 추출하는 방법을 사용한다. 인공 신경망의 이미지 분류에 사용되는 Convolution Neural Network (CNN) 구조 중 일반적으로 사용되는 Residual Learning Network (ResNet)와 Visual Geometry Group (VGG)를 사용하고자 한다. AI Hub에서 제공하는 한국인 얼굴 데이터 6만 장을 사용하며, 표정을 분류한 다. 분류하는 네 가지 표정은 슬픔, 행복, 분노, 그리고 무표정이다. 가장자리 검출 필터는 라플라시안 필터, 소벨 필터 및 샤르 필터, 캐니 가장자리 검출 필터를 사용한다. 그 후, 전처리 된 이미지를 모델에 학습하고, 테스트 데 이터에서 각 표정별 예측 정확도가 가장 높은 필터를 확인하고, 최종적으로 미세표정을 분류한 결과를 조합해 인 간의 심리를 파악하는 시스템을 제안하고자 한다.

In this paper, we use a method of extracting facial features by applying edge detection filters to images, rather than a method of classifying conventional facial expressions using facial landmarks. Among the Convolution Neural Network (CNN) structures used for image classification of artificial neural networks, we intend to use the commonly used ResNet and Visual Geometry Group (VGG). It uses 60,000 Korean face data provided by AI Hub and classifies facial expressions. The four expressions that classify are sad, happy, angry, and neutral. The edge detection filter uses a Laplacian filter, a Sobel filter and a Scharr filter, and a Canny edge detection filter. After that, we propose a system that learns preprocessing images from the model, identifies the filter with the highest prediction accuracy for each expression in the test data, and finally combines the results of classifying fine expressions to grasp human psychology.

본 연구는 간호학생의 팀기반학습(TBL)에서 공유리더십이 팀성과 및 팀몰입에 미치는 영향과 이들의 관계 에서 팀신뢰의 매개효과를 확인하기 위하여 시도되었다. 연구대상은 2016년 1학기 6주 동안 팀기반학습에 참여한 99 명의 간호대학생이었으며, 2016년 6월 자기기입식 설문지를 통해 자료를 수집하였다. 연구 결과, 공유리더십과 팀성 과, 팀몰입, 팀신뢰 간 모두 유의한 상관관계가 있는 것으로 나타났으며, 공유리더십은 팀성과와 팀몰입에 긍정적 영 향을 주는 것으로 나타났다. 또 팀신뢰는 공유리더십과 팀몰입과의 관계를 부분 매개하고 있는 것으로 나타났다. 본 연구를 통하여 팀기반학습에서 공유리더십은 팀성과와 팀몰입을 향상시키는데 매우 중요한 요인이며, 팀신뢰를 통하 여 팀몰입에 미치는 영향이 달라질 수 있다는 것을 알 수 있었다. 따라서 대학 내 팀기반학습 시 팀효과성을 높이기 위해서는 공유리더십과 팀신뢰를 높여야할 필요가 있다.

This study was to identify the effects of shared leadership on team performance and team commitment of Team-Based Learning(TBL) in nursing students and verify the mediating effects of team trust on the relationships between shared leadership and team performance and team commitment. Data were collected via survey from 99 nursing students to participate in TBL for 6weeks, in June 2016. The major findings were as follows: Shared leadership had a significant correlation with team performance and team commitment and team trust. Shared leadership had a positive effect on team performance and team commitment. Team trust had a partial mediating effect in the relationship between shared leadership and team commitment. The results of study show that shared leadership of TBL is very important for high team performance and team commitment and shared leadership make effect on team commitment under the mediation of team trust. Therefore, it is necessary to increase shared leadership and team trust for high team effectiveness in TBL.