Earticle

본 연구는 영화 개봉 전 활용 가능한 정적 데이터를 기반으로 AI 및 기계 학습(Machine Learning, ML) 기법을 활용한 영화 흥행 예측 모델을 제안하고, 이를 기존 베이스라인 모델과 비교하여 성능을 평가하였다. 2013년부터 2024년까지 수집된 KOBIS 데이터를 활용하여 배우, 감독, 배급사, 장르별 기술 통계와 상호작용 파생변수를 생성하였으며, RFECV(Recursive Feature Elimination with Cross- Validation)를 통해 주요 변수를 선별하였다. 또한, AutoML 기법을 적용하여 데이터셋에 최적화된 모델을 학습하고, 네이버 검색어 트렌드 데이터를 추가적으로 반영하여 예측 성능을 향상시켰다. 실험 결과, 최종 제안모델은 베이스라인 모델 및 기존 연구모델을 뛰어넘는 성능을 보였으며, F1-Macro 및 AHPR-Bingo를 활용한 평가 척도에서 높은 정확도를 기록하였다. 특히, 상호작용 파생변수는 모델 성능 개선에 기여하는 주요 변수로 나타났으며, 정적 데이터를 활용한 예측 모델이 영화 산업 내 의사결정 과정에서 유의미한 도구로 작용할 수 있음을 입증하였다. 본 연구는 정적 데이터만을 활용한 영화 흥행 예측 모델을 제안함으로써 학술적 및 실무적 기여를 제공한다. 연구 결과는 영화 제작사와 배급사가 개봉 전 전략적 의사결정을 지원하는 데 유용한 프레임워크를 제공하며, 향후 연구는 동적 데이터를 포함한 확장된 모델 개발 및 다양한 데이터셋 적용을 통해 본 연구를 보완할 수 있을 것으로 기대된다.

This study proposes a box office prediction model utilizing AI and machine learning (ML) techniques based on static data available before a film’s release and evaluates its performance by comparing it with existing baseline models. Using KOBIS data collected from 2013 to 2024, we generated descriptive statistics and interaction-derived variables for actors, directors, distributors, and genres. Key features were selected using Recursive Feature Elimination with Cross-Validation (RFECV). Additionally, an AutoML approach was applied to train an optimized model for the dataset, and Naver search trend data was incorporated to enhance predictive performance. Experimental results demonstrated that the proposed model outperformed both baseline and existing research models, achieving high accuracy based on F1-Macro and AHPR-Bingo evaluation metrics. Notably, interaction-derived variables were identified as key contributors to model performance improvement, confirming that prediction models based on static data can serve as valuable tools in the decision-making processes of the film industry. This study makes both academic and practical contributions by presenting a box office prediction model solely based on static data. The findings provide a useful framework for film producers and distributors to support strategic decision-making before release. Future research can extend this study by incorporating dynamic data and applying the model to diverse datasets for further validation and enhancement.

Recently, QUIC for the secure and faster connections has standardized but it is unclear that QUIC can cope with website fingerprinting (WF), a technique to infer visited websites from network traffic, since most existing efforts targeted TCP-induced traffic. To this end, we propose a novel QUIC WF technique based on Automated Machine Learning (AutoML). In our approach, we revisit traffic features appeared in literature, but relies on an AutoML framework to achieve best practice without manual intervention. Through experiments, we show that our technique outperforms state-of-the-art WF techniques with an F1-score of 99.79% and a 20-precision of 92.60%.

This study investigates the use of machine learning techniques to estimate the value of studio apartments (Officetel), which are increasingly important as combined office and residential spaces for single-person households and freelancers. It aims to identify key variables affecting studio apartment prices and preprocess them for accurate predictions. Transaction data from studio apartments are used to compare the predictive performance of four methods: Multiple Regression Analysis, Random Forest, XG Boosting, and Deep Learning. The study seeks to determine the best-performing models for price estimation and aims to develop predictive models applicable to various real estate types and regions.

하천의 과다한 조류 발생은 취수원 및 수생태 환경에 좋지 않은 영향을 줄 수 있으며 이에 대한 지속적인 관리가 중요하다. 본 연구에서는 낙동강 유입 지류 하천에서 조류 발생의 정량적 지표인 chlorophyll-a 농도를 예측하는 자동 머신러닝 모형 (AutoML: automated machine learning)을 구축하였다. AutoML은 머신러닝 모형의 구축을 위한 데이터 전처리, 모형 선정 및 최적화 과정의 편의성을 높여 상대적으로 용이한 모형 구축을 가능하게 하는 장점이 있으며, 본 연구에서는 AutoGluon을 이용하여 AutoML을 구축하고 기존에 널리 사용되는 머신러닝 모형인 random forest 및 XGBoost모형과 그 성능을 비교하였다. 모형 성능의 비교는 모형 성능의 평가에 활용되는 정량 지표인 Nash-Sutcliffe coefficient of efficiency (NSE), root mean squared error (RMSE) 및 RMSE-standard deviation ratio (RSR)를 활용하였다. 분석결과 AutoGloun, RF, XGB 세가지 모형의 RSR 값이 각각 0.564, 0.752, 0.811로 AutoGluon이 가장 우수한 성능을 보이는 것으로 확인되었다. 또한 AutoGluon 모형의 구축에 사용된 각 변수가 모형의 성능에 미치는 상대적 중요도인 feature importance를 확인하여 중요도가 낮은 변수부터 순차적으로 변수를 제거하면서 성능의 변화를 비교하였다. 분석결과 RSR이 0.542–0.579의 범위를 보여 입력변수가 제한적인 경우에도 일정 수준 이상의 안정적인 성능이 확보될 수 있음을 확인하였다.

Excessive algal blooms in rivers can have negative impacts on water resources and aquatic ecosystems. Therefore, continuous management of algal bloom is essential. In this study, an automated machine learning (AutoML) model was developed to predict chlorophyll-a concentrations. AutoML has the advantage of simplifying the machine learning model development process by streamlining data preprocessing, model selection, and optimization, thus making model construction relatively easier. In this study, AutoGluon was used to implement AutoML, and its performance was compared with that of widely used machine learning models (i.g. Random Forest and XGBoost). Model performance was evaluated using three quantitative metrics: Nash–Sutcliffe Efficiency (NSE), Root Mean Squared Error (RMSE), and the Root Mean Squared Error–Observation Standard Deviation Ratio (RSR). The analysis showed that the RSR values for AutoGluon, RF, and XGB models were 0.564, 0.752, and 0.811, respectively, indicating that AutoGluon demonstrated the best performance. Additionally, the relative importance of the input features used in the development of the AutoGluon model was explored. Features were sequentially removed based on their importance ranking to assess the impact on model performance. The results showed that the RSR ranged from 0.542 to 0.579, verifying that the model maintained a stable performance even when input variables were limited.

본 논문에서는 기존의 인포그래픽 제작방식을 개선하기 위하여 기계학습 기반의 인포그래픽 자동 추천 시스템을 제안하였다. 이 시스템은 복수의 인포그래픽 이미지를 기계학습하는 부분과 사용자의 기초자료 입력만으로 인포그래픽 을 인공지능으로 자동 추천하는 부분으로 구성된다. 추천된 인포그랙픽은 라이브러리 형태로 제공되고, 드래그 & 드롭 방식으로 추가적인 자료의 입력이 가능하게 된다. 또한, 입력한 자료의 크기에 따라 인포그래픽 이미지가 동적으로 조절 되도록 설계하였다. 기계학습 기반의 인포그래픽 자동 추천 과정을 분석한 결과 레이아웃과 키워드에 대한 일치 성공율 은 매우 높고, 타입에 대한 일치 성공률은 다소 낮게 나타났다. 추후 인포그래픽 부분별 이미지 타입에 대한 일치 성공 률을 향상시키기 위한 연구가 필요할 것이다.

In this paper, a machine learning-based automatic infographic recommendation system is proposed to improve the existing infographic production method. This system consists of a part that machine learning multiple infographic images and a part that automatically recommends infographics with artificial intelligence only by inputting basic data from the user. The recommended infographics are provided in the form of a library, and additional data can be input by drag & drop method. In addition, the infographic image is designed to be dynamically adjusted according to the size of the input data. As a result of analyzing the machine learning-based automatic infographic recommendation process, the matching success rate for layout and keyword was very high, and the matching success rate for type was rather low. In the future, a study to improve the matching success rate for the image type for each part of the infographic will be needed.

본 연구의 목적은 다양한 기계 학습 기법들이 서민 주거지로서 중요한 역할을 맡고 있으면서도 부동산 관련 연구 분야에서 소외되어온 다세대 주택의 가격 예측에 활용되는 과정과 각 기법의 예측 결과를 분석하고 조 사하는 것이다. 더불어 특정 지역 내 다세대 주택 가격 형성에 영향을 미치는 주요 속성 변수를 추출하고 전처리 하는 과정을 진행하였다. 서울시 마포구의 2020년부터 2022년까지 수집된 다세대 주택 실제 매매 데이터를 사례 로 MRA(다중 회귀 분석), RF(랜덤 포레스트), XG Boosting(XG 부스팅), Deep Learning(딥러닝)과 같은 4가지 기계 학습 방법의 예측 능력을 비교 분석하였으며, 다세대 주택의 가격을 평가하고 산정하는 작업에 있어 우수한 예측 성능을 가진 기계 학습 기법을 식별하고자 하였다. 본 연구 결과는 특정 지역의 데이터 셋을 구축하여 실험 적인 결과를 얻은 것이지만, 향후 다양한 지역 및 유형의 부동산 매물에도 적용 가능한 예측 모델을 개발하고자 시도하였다.

The aim of this study is to explore the application processes and analyze the predictive outcomes of various machine learning techniques in predicting multi-unit housing prices. Additionally, we conducted the extraction and preprocessing of key attribute variables influencing multi-unit housing prices in a specific region. Using actual transaction data from Mapo-gu, Seoul, spanning from 2020 to 2022, we compare the predictive capabilities of four machine learning methods: Multiple Regression Analysis(MRA), Random Forest(RF), XG Boosting, and Deep Learning. Our goal is to identify a machine learning technique with superior predictive performance for evaluating and estimating multi-unit housing prices. While the study's findings are based on experimental results from a specific regional dataset, we strive to develop a predictive model applicable to diverse regions and various real estate listings.

기계학습을 통한 문제 해결을 위한 단계에서 가장 중요한 부분 중의 하나가 데이터셋을 구축하는 것이다. 기존의 데 이터셋 구축 방법은 트레이닝 과정에서 소량의 이미지에 변화를 주어 데이터를 부풀리는 명령어를 사용하는 것과 ‘google 이미지 검색’에서 이미지 크롤링을 사용하는 것이 대부분이다. 하지만 이러한 방법들은 데이터로서 이미지 의 신뢰성이 보장되어 있지 않는 경우가 자주 발생한다. 기계학습 중 딥러닝 트레이닝을 위해서는 신뢰성이 보장된 데이터셋이 구축되어야 한다. 이에 따라 본 연구에서는 신뢰성을 보장받을 수 있는 데이터셋을 구축하기 위한 도구 를 설계 및 구현한다. 적은 노력으로 큰 데이터셋을 구축할 수 있는 방안을 고려하고, 그 방안을 통해 구축된 데이터 셋의 신뢰성을 테스트한다. 신뢰성을 보증하기 위한 예시로 꽃 이미지 데이터셋을 구축하여 ‘이미지 검색 꽃 사전’에 적용한다. 적용 결과로, 사용자는 딥러닝 트레이닝 과정에서 필요한 데이터셋을 보다 쉽고 경제적으로 구축할 수 있 고, 구축된 데이터셋을 통한 트레이닝에서의 신뢰성과 효율성이 있음을 보인다. 또한 본 연구의 아이디어를 기본으 로 활용한다면, 데이터셋을 구축하고자 하는 대상의 특성이 변하더라도 대량의 이미지 데이터셋을 생산할 수 있을 것이다.

One of the most important steps for solving problems through machine learning is to build a data set. Most previous methods of the data set generation are using commands to inflate data by changing small amounts of images during training, and image crawling in Google Image Search. However, these methods often do not guarantee the reliability of the image as data. For deeplearning training among methods of machine learning, a reliable data set must be established. Accordingly, in this study, a novel tool is designed and implemented to build a data set that can ensure reliability. We propose ways to build large data sets with little effort, and test the reliability of data sets built through them. As an example to ensure reliability, a set of flower image data is established and applied to ‘image search flower dictionary’. As a result, users can more easily and cost-effectively build the data sets needed in the deep-learning training process, and ensure reliability and efficiency in training through the generated data set. In addition, the underlying use of the ideas proposed in this study will enable the production of large sets of image data, even if the characteristics of the target to build the data set will be changed.

기존에 사용되던 방화벽들은 기본적으로 정책을 수동적으로 입력해 주는 방식으로 되어 있어 공격이 오는 즉시 대응하기 쉽지 않다. 왜냐하면 전문 보안 관리자가 이를 분석하고 해당 공격에 대한 방어 정책을 입력해 주어야하기 때문이다. 또한, 기존 방화벽 정책은 공격을 막기 위해 정상 접속까지 차단하는 경우가 많다. 패킷 자체는 정상적이지만 유입량이 많아 서비스 거부를 발생시키는 공격이 많기 때문이다. 본 논문에서는 방어 정책을 입력하는 부분을 인공지능 으로 대체하여 정책을 자동으로 생성하고, 정상 접속 학습을 통해 생성된 화이트리스트 정책으로 정상 접속은 가능하면 서 Flooding, Spoofing, Scanning과 같은 공격만을 차단하는 방법을 제안한다.

Conventional firewalls cannot cope with attacks immediately. It is because security professionals or administrators need to analyze them and enter relevant policies to the firewalls. In addition, those policies may often block even normal accesses. Even though the packet themselves are normal, there exist many attacks that cause denial of service due to the inflow of a large amount of those packets. In this paper, we propose a method to block attacks such as Flooding, Spoofing and Scanning while allowing normal accesses based on whitelist policies which are automatedly generated by learning normal access patterns.

도시의 인구증가 및 고밀화에 따라 기존 지하구조물에 인접하여 대심도 굴착 공사 물량이 증가하는 추세인 것 으로 분석되었다. 현재 지하구조물 및 궤도는 외부요인에 의해 지하구조물의 손상이 다수 발생되는 실정이며 터널 내 의 계측결과로 원인을 분석하여 예방차원이 아닌 사후처리에 대해서 측정을 하고 있는 실정이다. 본 연구의 목적은 공용중인 도시철도 선로와 인접한 굴착공사에 따른 구조물의 변형에 미치는 영향을 분석하는 것이다. 또한 외적 요인 으로 인해 지하구조물 및 궤도 손상 및 파괴가 발생하기 전 구조물의 변위를 머신러닝 기법을 통해 구조물의 안전성 을 평가하고자 한다. 분석결과, 분석한 데이터세트에서 구조물관리기준치에 도달하는 시간을 예측하기에 적합한 모델 은 다항회귀 머신러닝 알고리즘인 것으로 분석되었다. 그러나 본 연구에서 적용한 자동화계측 데이터에 한정될 수 있 으므로 추가적으로 구조물 조건의 다양성과 데이터양을 늘리는 향후 연구가 필요하다.

It was analyzed that the volume of deep excavation works adjacent to existing underground structures is increasing according to the population growth and density of cities. Currently, many underground structures and tracks are damaged by external factors, and the cause is analyzed based on the measurement results in the tunnel, and measurements are being made for post-processing, not for prevention. The purpose of this study is to analyze the effect on the deformation of the structure due to the excavation work adjacent to the urban railway track in use. In addition, the safety of structures is evaluated through machine learning techniques for displacement of structures before damage and destruction of underground structures and tracks due to external factors. As a result of the analysis, it was analyzed that the model suitable for predicting the structure management standard value time in the analyzed dataset was a polynomial regression machine. Since it may be limited to the data applied in this study, future research is needed to increase the diversity of structural conditions and the amount of data.

본 연구는 북극의 해빙표면온도(ice surface temperature, IST)를 자동화된 기계 학습(automated machine learning, AutoML) 기반으로 산출하였다. AutoML 기반 IST는 상관관계(correlation coefficient, R) 0.97, 평균 제곱근 오차(root mean squared error, RMSE) 2.51K로 산출되었다. 심층신경망(deep neural network, DNN) 모델과 비교하여 AutoML IST는 Moderate Resolution Imaging Spectroradiometer (MODIS) IST 및 ice mass balance (IMB) buoy IST와의 검증 결과에서 좋은 정확도를 보인다. 이는 어려운 극지방 조건에서 IST 추정 정확도를 향상시키는 AutoML의 효과를 강조한다.

This study utilized automated machine learning (AutoML) to calculate Arctic ice surface temperature (IST). AutoML-derived IST exhibited a strong correlation coefficient (R) of 0.97 and a root mean squared error (RMSE) of 2.51K. Comparative analysis with deep neural network (DNN) models revealed that AutoML IST demonstrated good accuracy, particularly when compared to Moderate Resolution Imaging Spectroradiometer (MODIS) IST and ice mass balance (IMB) buoy IST. These findings underscore the effectiveness of AutoML in enhancing IST estimation accuracy under challenging polar conditions.

Autism spectrum disorder (ASD) is a developmental disorder associated with cognitive and neurobehavioral disorders. It affects the person's behavior and performance. Autism affects verbal and non-verbal communication in social interactions. Early screening and diagnosis of ASD are essential and helpful for early educational planning and treatment, the provision of family support, and for providing appropriate medical support for the child on time. Thus, developing automated methods for diagnosing ASD is becoming an essential need. Herein, we investigate using various machine learning methods to build predictive models for diagnosing ASD in children using facial images. To achieve this, we used an autistic children dataset containing 2936 facial images of children with autism and typical children. In application, we used classical machine learning methods, such as support vector machine and random forest. In addition to using deep-learning methods, we used a state-of-the-art method, that is, automated machine learning (AutoML). We compared the results obtained from the existing techniques. Consequently, we obtained that AutoML achieved the highest performance of approximately 96% accuracy via the Hyperpot and tree-based pipeline optimization tool optimization. Furthermore, AutoML methods enabled us to easily find the best parameter settings without any human efforts for feature engineering.

Objectives: This study aimed to develop and validate machine learning (ML) models using H2O-AutoML, an automated ML program, for predicting medication-related osteonecrosis of the jaw (MRONJ) in patients with osteoporosis undergoing tooth extraction or implantation. Patients and Methods: We conducted a retrospective chart review of 340 patients who visited Dankook University Dental Hospital between January 2019 and June 2022 who met the following inclusion criteria: female, age ≥55 years, osteoporosis treated with antiresorptive therapy, and recent dental extraction or implantation. We considered medication administration and duration, demographics, and systemic factors (age and medical history). Local factors, such as surgical method, number of operated teeth, and operation area, were also included. Six algorithms were used to generate the MRONJ prediction model. Results: Gradient boosting demonstrated the best diagnostic accuracy, with an area under the receiver operating characteristic curve (AUC) of 0.8283. Validation with the test dataset yielded a stable AUC of 0.7526. Variable importance analysis identified duration of medication as the most important variable, followed by age, number of teeth operated, and operation site. Conclusion: ML models can help predict MRONJ occurrence in patients with osteoporosis undergoing tooth extraction or implantation based on questionnaire data acquired at the first visit.

Increased use of software and complexity of software functions, as well as shortened software quality evaluation periods, have increased the importance and necessity for automation of software testing. Automating software testing by using machine learning not only minimizes errors in manual testing, but also allows a speedier evaluation. Research on machine learning in automated software testing has so far focused on solving special problems with algorithms, leading to difficulties for the software developers and testers, in applying machine learning to software testing automation. This paper, proposes a new machine learning framework for software testing automation through related studies. To maximize the performance of software testing, we analyzed and categorized the machine learning algorithms applicable to each software test phase, including the diverse data that can be used in the algorithms. We believe that our framework allows software developers or testers to choose a machine learning algorithm suitable for their purpose.

This article presents a study on the state-of-the-art methods for automated radioactive material detection and identification, using gamma-ray spectra and modern machine learning methods. The recent developments inspired this in deep learning algorithms, and the proposed method provided better performance than the current state-of-the-art models. Machine learning models such as: fully connected, recurrent, convolutional, and gradient boosted decision trees, are applied under a wide variety of testing conditions, and their advantage and disadvantage are discussed. Furthermore, a hybrid model is developed by combining the fully-connected and convolutional neural network, which shows the best performance among the different machine learning models. These improvements are represented by the model's test performance metric (i.e., F1 score) of 93.33% with an improvement of 2%-12% than the state-of-the-art model at various conditions. The experimental results show that fusion of classical neural networks and modern deep learning architecture is a suitable choice for interpreting gamma spectra data where real-time and remote detection is necessary.

Purpose: This study aimed to assess the usefulness of machine learning and automation techniques to match pairs of panoramic radiographs for personal identification. Materials and Methods: Two hundred panoramic radiographs from 100 patients (50 males and 50 females) were randomly selected from a private radiological service database. Initially, 14 linear and angular measurements of the radiographs were made by an expert. Eight ratio indices derived from the original measurements were applied to a statistical algorithm to match radiographs from the same patients, simulating a semi-automated personal identification process. Subsequently, measurements were automatically generated using a deep neural network for image recognition, simulating a fully automated personal identification process. Results: Approximately 85% of the radiographs were correctly matched by the automated personal identification process. In a limited number of cases, the image recognition algorithm identified 2 potential matches for the same individual. No statistically significant differences were found between measurements performed by the expert on panoramic radiographs from the same patients. Conclusion: Personal identification might be performed with the aid of image recognition algorithms and machine learning techniques. This approach will likely facilitate the complex task of personal identification by performing an initial screening of radiographs and matching ante-mortem and post-mortem images from the same individuals.

Background and Purpose Multiphase computed tomographic angiography (mCTA) provides time variant images of pial vasculature supplying brain in patients with acute ischemic stroke (AIS). To develop a machine learning (ML) technique to predict tissue perfusion and infarction from mCTA source images. Methods 284 patients with AIS were included from the Precise and Rapid assessment of collaterals using multi-phase CTA in the triage of patients with acute ischemic stroke for Intra-artery Therapy (Prove-IT) study. All patients had non-contrast computed tomography, mCTA, and computed tomographic perfusion (CTP) at baseline and follow-up magnetic resonance imaging/noncontrast- enhanced computed tomography. Of the 284 patient images, 140 patient images were randomly selected to train and validate three ML models to predict a pre-defined Tmax thresholded perfusion abnormality, core and penumbra on CTP. The remaining 144 patient images were used to test the ML models. The predicted perfusion, core and penumbra lesions from ML models were compared to CTP perfusion lesion and to follow-up infarct using Bland-Altman plots, concordance correlation coefficient (CCC), intra-class correlation coefficient (ICC), and Dice similarity coefficient. Results Mean difference between the mCTA predicted perfusion volume and CTP perfusion volume was 4.6 mL (limit of agreement [LoA], ?53 to 62.1 mL; P=0.56; CCC 0.63 [95% confidence interval [CI], 0.53 to 0.71; P<0.01], ICC 0.68 [95% CI, 0.58 to 0.78; P<0.001]). Mean difference between the mCTA predicted infarct and follow-up infarct in the 100 patients with acute reperfusion (modified thrombolysis in cerebral infarction [mTICI] 2b/2c/3) was 21.7 mL, while it was 3.4 mL in the 44 patients not achieving reperfusion (mTICI 0/1). Amongst reperfused subjects, CCC was 0.4 (95% CI, 0.15 to 0.55; P<0.01) and ICC was 0.42 (95% CI, 0.18 to 0.50; P<0.01); in non-reperfused subjects CCC was 0.52 (95% CI, 0.20 to 0.60; P<0.001) and ICC was 0.60 (95% CI, 0.37 to 0.76; P<0.001). No difference was observed between the mCTA and CTP predicted infarct volume in the test cohort (P=0.67). Conclusions A ML based mCTA model is able to predict brain tissue perfusion abnormality and follow-up infarction, comparable to CTP.

The outbreak of COVID-19 has dramatically changed the dynamics of learning environment. Various researches have examined the implications and impacts of such changes and the results were met with mixed findings. Although the learners inevitably preferred face to face interaction and direct communication, under the circumstances they were willing to continue with the current devices if multiple modes of communication devices could be added to the current environment. The paper aimed to utilize the use of Automated Translating Machines (ATM), such as Naver Papago and Google Translator, to learners and to examine whether such devices could further aid the learners under such circumstances. Results showed that the high proficiency group only used the ATMs for reference purposes and such devices did not yield meaningful findings even with certain target vocabularies. However, low proficiency group learners were able to utilize the device to learn certain target vocabularies better than the conventional vocabulary learning method. This shows that ATMs can be utilized with the online-learning to teach certain target vocabularies to the low proficiency group learners.

지속 가능한 에너지인 태양광 발전은 전 세계에서 널리 활용하는 재생 에너지 원천 중 하나로 최근 효율적인 태양광 발전 시스템 운영을 위해 태양광 발전량을 정확하게 예측하기 위한 연구가 활발히 진행되고 있다. 태양광 발전량 예측 모델을 구성하기 위해서는 기상 및 대기 환경을 넘어 태양의 위치에 따른 일사량의 정보가 필수적이나 태양의 실시간 위치 정보를 입력 변수로 활용한 연구가 부족한 실정이다. 그리하여 본 논문에서는 시간과 태양광 발전소 위치를 기반으로 태양의 고도와 방위각을 실시간으로 계산하여 입력 변수로 사용하는 방식을 제안한다. 이를 위해 AutoML 기반의 다양한 기계학습 모델을 구성하여 태양광 발전율을 예측하고 그 성능을 비교 분석하였다. 실험 결과, 태양 위치 정보를 포함한 경우에 환경 변수만을 고려하였을 때보다 예측 성능이 크게 향상되었음을 확인할 수 있었으며, Extra Trees 모델의 경우 태양 위치 정보를 추가하였을 때 MAE(Mean Absolute Error)가 33.90 에서 22.38 까지 낮아지는 결과를 확인하였다.

The performance of machine learning algorithms significantly depends on how a configuration of hyperparameters is identified and how a neural network architecture is designed. However, this requires expert knowledge of relevant task domains and a prohibitive computation time. To optimize these two processes using minimal effort, many studies have investigated automated machine learning in recent years. This paper reviews the conventional random, grid, and Bayesian methods for hyperparameter optimization (HPO) and addresses its recent approaches, which speeds up the identification of the best set of hyperparameters. We further investigate existing neural architecture search (NAS) techniques based on evolutionary algorithms, reinforcement learning, and gradient derivatives and analyze their theoretical characteristics and performance results. Moreover, future research directions and challenges in HPO and NAS are described.