Earticle

본 논문에서는 기계학습과 관련된 다양한 사례들에 대한 연구를 바탕으로 기계학습 응용 및 학습 알고리즘의 성능 개선 방안을 제시한다. 이를 위해 기계학습 기법을 적용하여 결과를 얻어낸 문헌을 자료로 수집하고 학문 분야로 나누어 각 분야에서 적합한 기계학습 기법을 선택 및 추천하였다. 공학에서는 SVM, 의학에서는 의사결정나무, 그 외 분야에서는 SVM이 빈번한 이용 사례와 분류/예측의 측면에서 그 효용성을 보였다. 기계학습의 적용 사례 분석을 통해 응용 방안의 일반적 특성화를 꾀할 수 있었다. 적용 단계는 크게 3단계로 이루어진다. 첫째, 데이터 수집, 둘째, 알고리즘을 통한 데이터 학습, 셋째, 알고리즘에 대한 유의미성 테스트 이며, 각 단계에서의 알고리즘의 결합을 통해 성능을 향상시킨다. 성능 개선 및 향상의 방법은 다중 기계학습 구조 모델링과 +α 기계학습 구조 모델링 등으로 분류한다.

This paper aims to present the way to bring about significant results through performance improvement of learning algorithm in the research applying to machine learning. Research papers showing the results from machine learning methods were collected as data for this case study. In addition, suitable machine learning methods for each field were selected and suggested in this paper. As a result, SVM for engineering, decision-making tree algorithm for medical science, and SVM for other fields showed their efficiency in terms of their frequent use cases and classification/prediction. By analyzing cases of machine learning application, general characterization of application plans is drawn. Machine learning application has three steps: (1) data collection; (2) data learning through algorithm; and (3) significance test on algorithm. Performance is improved in each step by combining algorithm. Ways of performance improvement are classified as multiple machine learning structure modeling, +α machine learning structure modeling, and so forth.

Three-dimensional integrated circuits (3D ICs) experience die-to-die variations in addition to the already challenging within-die variations. This adds an additional design complexity and makes variation estimation and full-chip optimization even more challenging. In this paper, we show that the industry standard on-chip variation (AOCV) tables cannot be applied directly to 3D paths that are spanning multiple dies. We develop a new machine learning-based model and methodology for an accurate variation estimation of logic paths in 3D designs. Our model makes use of key parameters extracted from existing GDSII 3D IC design and sign-off simulation database. Thus, it requires no runtime overhead when compared to AOCV analysis while achieving an average accuracy of 90% in variation evaluation. By using our model in a full-chip variation-aware 3D IC physical design flow, we obtain up to 16% improvement in critical path delay under variations, which is verified with detailed Monte Carlo simulations.

본 연구는 최신 전환학습(Transfer learning) 텍스트 모델링 기법인 BERTopic을 통해 총 38,312개의 코로나19 백신 관련 기사를 분석함으로써 22개의 주요 토픽을 생성하였고, 그중에 가장 많이 언급된 토픽들은 코로나 확진자 추세, 백신 접종, 치료제 개발, 변이와 집단감 염, 등등의 순이었다. 시계열 요소를 도입한 BERTopic dynamic topic model을 통해 살펴본 토픽 의 변화는 상대적으로 예방접종이 잘 진행되고 있던 시기에는 백신 접종에 대한 보도가 줄어들고 델타 변이가 출연한 시점부터 오히려 백신 보도의 비율이 늘어나는 양상을 보였다. 시계열 특성 을 회귀분석에 포함한 방법으로 모델을 추정하였을 때, 단순히 코로나 확진자 수를 보도하거나 백신 개발을 강조하는 보도는 오히려 백신 접종자 수를 감소시켰고, 반면에 해외여행 등 백신을 접종함으로써 받을 수 있는 이익을 강조하는 보도가 백신 접종자 수의 증가와 관련이 있었다. 본 연구는 언론보도와 공중의 의견 간의 관계를 밝히는 기존 연구들의 의의를 넘어 특정 언론보도의 주제가 공중의 행동(백신 접종)에 영향을 미친다는 것을 밝혀 향후 연구자, 광고/홍보 전문가 그 리고 정책 결정자가 보건 위기 상황에 대응할 방안을 마련했다는 데 그 의의가 있다.

Utilizing cutting-edge text modeling techniques such as BERTopic employing a transfer learning approach, the current study investigates a total of 38,312 news articles regarding COVID-19 vaccination. We found that COVID-19 confirmed cases, vaccination efforts, the development of COVID-19 treatments, variations, and mass infections are the prominent topics among the 22 topics that we extracted. Considering the time series dynamics, we observed a decrease in topics related to vaccination as the actual vaccination rate steadily increased, whereas it increased after the appearance of the Delta variant. Regarding the relationship between topics and actual vaccination cases, we identified a negative association when newspapers reported the number of confirmed cases. However, emphasizing the benefits of vaccination, such as the availability of overseas travel, led to an increase in vaccination rates. These findings offer novel insights for theory developers, advertising/public relations experts, and policymakers dealing with public health crises.

Post-marketing surveillance of adverse events in real-world settings is a key process in drug safety management, and signal detection through spontaneous reporting systems provides critical evidence for regulatoryscientific decision-making. This study analyzed GLP-1 Receptor Agonists (GLP-1 RAs) class-related adverse event signals using the FDA Adverse Event Reporting System (FAERS) and identified factors influencing their formation. FAERS data from the second quarter of 2021 to the third quarter of 2025 were analyzed using disproportionality metrics (PRR, ROR, IC). The exposure group included all GLP-1 RA class drugs, and the comparator group comprised non-GLP-1 receptor agonists used for diabetes and obesity. Detected signals were categorized by MedDRA System Organ Class (SOC), and machine learning models (Random Forest (RF), XGBoost) were applied to identify predictors and contextual patterns of gastrointestinal (GI) adverse events. Signal detection results showed that GI disorders were frequently reported in the exposure group (a=211,797), with strong signals (ROR 2.94 [95% CI: 2.92–2.97]. The XGBoost and RF models achieved a high AUROC of 0.926 and 0.929, respectively, identifying patient age (45-64 years), seriousness, and reporter type as major predictors of GI-related signals. This suggests that GI adverse event reporting may be associated with specific reporting contexts and demographics, such as ‘patient age (45-64 years)’ and ‘seriousness’. The model's input features could predict whether a report was for a GI event with 92.9% accuracy. This study demonstrates the practical applicability of regulatory-scientific analytics, showing that combining machine learning with spontaneous reporting data enhances the accuracy and efficiency of post-marketing signal detection and provides stronger evidence for pharmacovigilance and risk management.

온라인 C2C 중고거래에 대한 수요가 증가하고 있으나 물품을 보내지 않거나 명시한 것과 다른 물건을 보내는 방식으로 부당한 금전적 이득을 챙기려는 사기 행위자들의 수도 증가하고 있다. 본 연구는 이러한 사기를 미연에 방지하기 위한 머신 러닝 방법을 이용한 사기 탐지 모델을 구축하였다. 이를 위해 대표적 C2C 중고거래 플랫폼인 중고나라에서 145,536건의 거래 게시글을 수집하였다. 이후 이들 게시글에서 토픽 모델링 기법을 이용하여 상품 설명 내용의 주제를 추출하였으며, 상품 설명의 언어적 특성, 준언어적 특성, 상품의 특성, 게시글의 포스팅 특성, 구매자 특성, 거래 특성들을 추출하였다. 이를 XGBoost 방법에 기반한 머신 러닝 모델을 구축하여 사기 게시글을 탐지하였다. 분석 결과, 사기 게시글은 글 자체의 길이가 대체로 짧고, 제공하는 정보가 적고 상대적으로 구체적이지 않은 것으로 나타났으며 명사를 상대적으로 적게 쓰고 이미지도 사용하지 않거나 적게 사용하는 글이 대부분인 것으로 나타났다. 또한 상대적으로 숫자와 공백의 비율이 높게 나타났으며 정상 게시글의 경우 명사의 경우 상품의 정보, 동사의 경우 전달, 형용사의 경우는 행위와 관련된 단어들이 사용되었으나 사기 게시글은 뚜렷한 주제를 가지지 못하는 것으로 나타났다. 본 연구는 전화번호나 계좌번호를 사용한 기존의 방법과 달리 다양한 게시글의 특성으로 사기 여부를 탐지하는 모델을 구축했다는 점에서 학술적, 실무적 시사점을 가지고 있다.

As the transaction volume of the C2C second-hand market is growing, the number of frauds, which intend to earn unfair gains by sending products different from specified ones or not sending them to buyers, is also increasing. This study explores the model that can identify frauds in the online C2C second-hand market by examining the postings for transactions. For this goal, this study collected 145,536 field data from actual C2C second-hand market. Then, the model is built with the characteristics from postings such as the topic and the linguistic characteristics of the product description, and the characteristics of products, postings, sellers, and transactions. The constructed model is then trained by the machine learning algorithm XGBoost. The final analysis results show that fraudulent postings have less information, which is also less specific, fewer nouns and images, a higher ratio of the number and white space, and a shorter length than genuine postings do. Also, while the genuine postings are focused on the product information for nouns, delivery information for verbs, and actions for adjectives, the fraudulent postings did not show those characteristics. This study shows that the various features can be extracted from postings written in C2C second-hand transactions and be used to construct an effective model for frauds. The proposed model can be also considered and applied for the other C2C platforms. Overall, the model proposed in this study can be expected to have positive effects on suppressing and preventing fraudulent behavior in online C2C markets.

본 연구는 블록체인 기술도입에 따른 핵심 토픽을 강원경제 주요 산업인 헬스케어, 의료기 기, 관광 및 식품 산업별로 고찰하였다. 조선일보, 중앙일보, 동아일보 등 주요 일간지 및 강원일보에 게재된 신문기사에서 2016년 1월 1일부터 최근까지의 ‘블록체인’ 관련 기사를 크롤링하여 키워드 빈도분석, 기계학습 방식인 LDA 기반의 토픽 모델링, 토픽 네트워크 및 클러스터링 분석 등을 수행하였다. 토픽 모델링을 통해 강원경제 전략산업별 상위 토픽을 추출하였고, 토픽 네트워크 분석을 통해 토픽 간 거리를 측정하였고 토픽 클러스터링을 실시하여 토픽의 연결중심성과 토픽 간 연계성을 측정하였다. 분석 결과 발견 된 주요 사실은 다음과 같다. 첫째, 국내 블록체인 기술응용은 대체로 1세대에 큰 비중을 두고 있다. 둘째, 강원경제의 경우 관련 주제 간 연계성이 상대적으로 낮은 편이다. 셋째, 산업별 핵심 토픽이 상이하며 핵심 토픽의 연결중심성 및 토픽 간 연계성 수준이 상이하여 블록체인 기술 도입으로 인한 기대효과는 산업별로 상이한 투입-산출 경로를 통해 나타날 것으로 기대할 수 있다.

This paper examines main topics accompanied by applying blockchain technology to the core industries of Gangwon economy such as healthcare, medical devices, tourism and food industries. Raw data has been collected from crawling texts concerning ‘blockchain’ in the main daily newspapers such as Chosun-, Chungang-, Donga- and Gangwon Daily Newspaper, covering the research period of 1. Jan. 2016 to the current. Analytic tools cover the key words frequency analysis, machine learning LDA-based topic modelling, topic network & clustering analysis. The main findings are: 1) in the domestic level the application of blockchain technology gives weight to the first generation, 2) in the case of Gangwon economy the connectedness between topics is relatively low, 3) core topics of every single industries, the degree centrality of them and the connectedness between them are different. Therefore, it is possible to expect that the effect of adopting blockchain technology will occur through different input-output processes by industries.

본 논문은 통합전자전 상황에서 전자정보 수집신호의 변수를 분석하여 전자파 신호를 발산하는 레이더 및 위협체계를 역변환으로 모델링하는 기법을 다루고자 한다. 전자전 수신기가 위협체로부터 수신한 변수 중 가장 중요한 속성을 1-rule 알고리즘을 통하여 파악하였으며, 베이지안 분류 알고리즘, 의사결정 트리 학습 알고리즘 및 인공 신경망알고리즘을 사용하여 위협체를 추정할 수 있는 프로토타입을 구축하였다. 다양한 통합전자전 상황에서 생성한 규칙을 기반으로 역변환 모델을 구축하고 새로운 데이터가 주어지면 모델링한 위협체로 분류한다. 실험에서 베이지안 분류 알고리즘, 의사결정 트리 학습 알고리즘 및 인공 신경망 알고리즘을 각각 적용하여 학습 곡선을 생성하였으며, 학습 곡선에서 최적의 성능을 나타내는 역변환 모델을 실험적으로 정립하였다.

This paper proposes a method to model the reverse extrapolation of RF threats through the attributes representing RF threats in an integrated electronic warfare situation. Using 1-rule algorithm, the most important attribute figuring out the threat could be assured among the attributes which a receiver collected. And also, we build a prototype which estimates threats using three machine learning algorithms, i.e., naive Bayesian classifier, decision tree, and artificial neural network. In various integrated electronic warfare situations, we tested our reverse extrapolation model based on the outputs of machine learning algorithms, where new test data were given to identify a threat. We also generated learning curves for each algorithm, and experimentally decided their optimal reverse extrapolation models. To verify the prototype of our model compiled, we evaluated the accuracy of reverse extrapolation models and the resulting performance showed more than 99% of accuracy.

In order to reduce environmental pollution, it is necessary to increase the recycling rate of waste. For this, the separation of recyclables is of utmost importance. The paper conducted a study to automatically discriminate containers by material for beverage containers among recyclables. We developed an algorithm that automatically recognizes containers by four materials: metal, glass, plastic, and paper by measuring the vibration signal generated when the beverage container collides with the bottom plate of the collection box. The amplitude distribution, time series information, and frequency information of the vibration signal were used to extract the characteristics indicating the characteristic difference of the vibration signal for each material, and a classifier was developed through machine learning using these characteristics.

1인 창조기업에 대한 정부의 지원이 확대되고, 인공지능과 IT 기술의 발달로 1인 창조기업의 성장할 수 있는 환경이 좋아졌다. 이로 인해 1인 창조기업에 대한 사람들의 관심이 증가하고, 1인 창조기업 지원센터를 찾는 청년 창업가의 숫자도 크게 증가하고 있다. 본 연구는 비지도 기계학습과 베이지안 네트워크를 결합한 새로운 유형의 확률적 구조방정식 모형, 즉 PSEM을 사용하여 국내 1인 창조기업의 당기순이익에 영향을 주는 요인을 분석하고자 한다. 아울러 해당 영향요인들이 국내 1인 창조기업의 당기순이익에 미치는 인과관계 값을 산출하기 위하여 What-If 분석을 실시하였다. 분석 결과 당기 순이익의 변화에 큰 영향을 받지 않는 요인들은 서비스 업종일 때 ‘초기 자본금’, ‘창업자특성’, ‘준비 기간’, ‘창업자 배경’, ‘회사 특성’으로 나타났다. 서비스 업종을 제외하였을 때는 ‘창업자 배경’, ‘인식’, ‘초기 자본금’, ‘회사 특성’, ‘창업자특성’으로 분석되었다. 당기순이익에 영향을 주는 요인은 ‘수익’, ‘종업원’, ‘판매’임을 발견하였다.

In the era of economic recession caused by covid-19 pandemic, government expanded various types of supports for one-man creative companies with a bid to promoting healthier ecosystem for start-ups. Furthermore, the 4th industrial revolution atmospheres emboldened by AI technologies have countenanced young entrepreneurs’attempts to launch their own one-man creative companies. However, despite positive trends with one-man creative companies like this, few studies exist investigating which factors affect its net profit. To fill the research void, this study proposes a novel method of probabilistic structural equation model (PSEM) combining unsupervised machine learning and Bayesian networks. Empirical results revealed that those influential factors affecting one-man creative companies’ net profit encompass ‘income’, ‘number of employees’, and ‘sales’. On the contrary, multiple factors initially deemed to be significant to the net profit were excluded, like “background of the CEO”, “characteristics of the company”, “initial fund”, ‘preparation’, and “characteristics of the CEO” in service company. The others company’s results showed that ‘background of the CEO’, ‘perception’, ‘initial fund’, ‘characteristics of the company’, and ‘characteristics of the CEO’ were not included in factors affected by changes of net profit.

In this paper, we propose a new hybrid intelligent modeling using context clustering and Extreme Learning Machine (ELM) mechanism. It has been a sensitive issue that the ELM mechanism assigns initial parameters randomly, despite of its superior performance. The proposed approach focuses on initial parameters determination of the modeling to improve the accuracy of the ELM mechanism, through removing randomness of assignment. To accomplish it, a context clustering based on Gaussian Mixture Model (GMM), considering a relationship between input-output spaces will be adopted to a Radial Basis Function Network (RBFN) of the ELM. In addition, the proposed approach will reduce the randomness of results from the original ELM. Simulations and the results show usefulness of the proposed approach with improved performance accuracy.

최근 부동산 시장에 대한 관심이 높다. 과거 주거환경으로만 여겨지던 부동산은 끊임없는 수요 증가로 안정적인 투자 대상으로 인식되고 있기 때문이다. 특히 국내 시장의 경우 인구 수의 감소에도 불구하고 1인 가구의 증가 및 대도시 로의 인구 유입이 가속화되며 수도권 중심으로 부동산 가격이 급격히 상승하고 현상이 나타나고 있다. 이에 미래 부동산 시장의 전망을 정확히 예측하는 것은 개인의 자산 관리 뿐 아니라 정부 정책 수립 등 사회 전반에 걸쳐 매우 중요한 사안이라고 할 수 있다. 본 논문에서는 머신러닝 기법을 활용해 과거 부동산 매매 데이터를 학습해 미래 부동산 시세를 예측하는 프로그램을 개발하였다. 한국감정원과 국토교통부에서 제공하는 대한민국 부동산 매매 시세 데이터를 활용하 였으며 지역별로 2022년도 평균 매매가 예측치를 제시한다. 개발된 프로그램은 오픈소스 형태로 공개하여 다양한 형태 로 활용될 수 있도록 하였다.

Recently, the real estate is of high interest. This is because real estate, which was considered only a residential environment in the past, is recognized as a stable investment target due to the ever-growing demand on it. In particular, in the case of the domestic market, despite the decrease in the number of people, the number of single-person households and the influx of people to large cities are accelerating, and real estate prices are rising sharply around the metropolitan area. Therefore, accurately predicting the prospects of the future real estate market becomes a very important issue not only for individual asset management but also for government policy establishment. In this paper, we developed a program to predict future real estate market prices by learning past real estate sales data using machine learning techniques. The data on the market price of real estate provided by the Korea Appraisal Board and the Ministry of Land, Infrastructure and Transport were used, and the average sales price forecast for 2022 by region is presented. The developed program is publicly available so that it could be used in various forms.

스마트 그리드는 전기 공급 업체와 소비자 간의 양방향 통신을 가능하게 하는 새로운 패러다임이다. 기존 방법은 단기 의존성을 처리하는 데만 유용하기 때문에 비선형 전력 수요 패턴의 모델링과 관련해서는 상당 부분 취약하다. 또한 기존 방법은 순전히 기록 데이터 기반이기 때문에 특성상 정적이고 스마트 미터기 오작동, 잡음과 같은 이상값이 상당 부분 포함되어 있어서 예측 성능이 매우 미흡하다. 이에 본 연구에서는 두 가지 기본 모델로서 하나는 모수 통계의 대표 적인 회귀 학습 모델과 다른 하나는 전형적인 비모수 통계적 접근이라 할 수 있는 KNN 학습 모델로 시작해서 스마트 미터 데이터와 같은 시계열 데이터의 경우에 샘플 데이터의 독립성이라는 측면을 고려하면, 오히려 비모수적 접근 방식 인 KNN 학습에 의한 성능이 본 연구의 실험 결과 확인된다.

Smart grid is a new paradigm that enables two-way communication between electricity suppliers and consumers. Existing methods are largely weak when it comes to modeling nonlinear power demand patterns, as they are only useful for handling short-term dependencies. Additionally, because existing methods are purely based on historical data, they are static in nature and contain a significant amount of outliers such as smart meter malfunctions and noise, so their prediction performance is very poor. Accordingly, in this study, we start with two basic models, one is a regression learning model that is representative of parametric statistics and the other is a KNN learning model that can be considered a typical non-parametric statistical approach. In the case of time series data such as smart meter data, we start with the independence of sample data. Considering this aspect, the performance of KNN learning, which is a non-parametric approach, is confirmed by the experimental results of this study

자동화된 스마트 그리드의 등장은 이러한 문제에 대응을 위한 필수적인 장치가 되고 있으며 스마트 그리드 기반 사회로의 진전을 가져오고 있다. 스마트 그리드는 전기 공급 업체와 소비자 간의 양방향 통신을 가능하게 하는 새로운 패러다임이다. 스마트 그리드는 전력 그리드를 보다 안정적이고 신뢰할 수 있으며 효율적이고 안전하게 만들기 위한 엔 지니어의 이니셔티브로 인해 등장했다. 스마트 그리드는 전력 소비자가 전력 사용에서 더 큰 역할을 할 수 있는 기회를 창출하고 전력을 현명하고 효율적으로 사용하도록 동기를 부여한다. 이에 본 연구에서는 기계 학습을 통한 전력 수요 관리에 중점을 둔다. 기계 학습을 사용한 수요 예측과 관련하여 현재 다양한 기계 학습 모델이 소개되어 적용되고 있는 데 이에 관한 체계적인 접근이 요구되고 있다. 특히 GP 학습 모델의 경우에 일반 소비 예측 및 데이터의 가시화와 관련 해서 다른 학습 모델보다 장점이 있지만, 스마트 미터 데이터의 예측과 관련해서는 데이터 독립성에 강한 영향을 받는다.

The emergence of the automated smart grid has become an essential device for responding to these problems and is bringing progress toward a smart grid-based society. Smart grid is a new paradigm that enables two-way communication between electricity suppliers and consumers. Smart grids have emerged due to engineers' initiatives to make the power grid more stable, reliable, efficient and safe. Smart grids create opportunities for electricity consumers to play a greater role in electricity use and motivate them to use electricity wisely and efficiently. Therefore, this study focuses on power demand management through machine learning. In relation to demand forecasting using machine learning, various machine learning models are currently introduced and applied, and a systematic approach is required. In particular, the GP learning model has advantages over other learning models in terms of general consumption prediction and data visualization, but is strongly influenced by data independence when it comes to prediction of smart meter data.

Silica nanoparticles, which have a broad range of sizes and specific surface features, have been used in many industrial applications. This study was conducted to synthesize monodispersed silica nanoparticles directly from tetraethyl orthosilicate (TEOS) with an alkaline catalyst (NH₃) based on the sol-gel process and the Stöber method. A central composite design (CCD) is used to build a second-order (quadratic) model for the response variables without requiring a complete three-level factorial experiment. The process was then optimized to achieve the minimum particle size with the lowest concentration of TEOS. Dynamic light scattering and scanning electron microscopy were used to analyze the size, dispersity, and morphology of the synthesized nanoparticles. After optimization, a confirmation test was carried out to evaluate the confidence level of the software prediction. The results revealed that the predicted optimization is consistent with experimental procedures, and the model is significant at the 95% confidence level.

In this study, machine learning (ML) techniques were used to model surveillance test data of nuclear power plants from an international database of the ASTM E10.02 committee. Regression modeling was conducted using various techniques, including Cubist, XGBoost, and a support vector machine. The root mean square deviation of each ML model for the baseline dataset was less than that of the ASTM E900-15 nonlinear regression model. With respect to the interpolation, the ML methods provided excellent predictions with relatively few computations when applied to the given data range. The effect of the explanatory variables on the transition temperature shift (TTS) for the ML methods was analyzed, and the trends were slightly different from those for the ASTM E900-15 model. ML methods showed some weakness in the extrapolation of the fluence in comparison to the ASTM E900-15, while the Cubist method achieved an extrapolation to a certain extent. To achieve a more reliable prediction of the TTS, it was confirmed that advanced techniques should be considered for extrapolation when applying ML modeling.

Aluminum-based composites are in high demand in industrial fields due to their light weight, high electrical conductivity, and corrosion resistance. Due to its unique advantages for composite fabrication, powder metallurgy is a crucial player in meeting this demand. However, the size and weight fraction of the reinforcement significantly influence the components' quality and performance. Understanding the correlation of these variables is crucial for building high-quality components. This study, therefore, investigated the correlations among various parameters?namely, milling time, reinforcement ratio, and size?that affect the composite’s physical and mechanical properties. An artificial neural network model was developed and showed the ability to correlate the processing parameters with the density, hardness, and tensile strength of Al2024-B4C composites. The predicted index of relative importance suggests that the milling time has the most substantial effect on fabricated components. This practical insight can be directly applied in the fabrication of high-quality Al2024-B4C composites.

The construction and materials sector is actively striving to mitigate the environmental consequences of cement production in concrete by integrating alternative and supplemental cementitious materials while reducing carbon emissions. Because of their pozzolanic reactions, natural pozzolans (NPs) have become prominent in this area. The aim of this research is to accurately predict the compressive strength of normal-weight concrete that contains NP by investigating the impact of several elements, including cement, NP content, water and aggregate quantity, and superplasticizer content. For doing this, the research examined data gathered from various sources, which led to the creation of a dataset consisting of 496 mix ratios with strengths. A comprehensive analysis was conducted using numerous advanced machine learning (ML) algorithms, including extreme gradient boosting (XGB), adaptive boosting (ADB), and bagging regressor (BAG), as well as hybrid ML techniques such as XGB-ADB and XGB-BAG. The purpose was to extensively examine the concrete mix materials and evaluate their influence on strength. The collected dataset was divided into two groups: training and testing. Statistical tests were conducted to ascertain the correlations between the input parameters and strength. Furthermore, the algorithms' performance was assessed using four separate statistical assessment criteria. The hybrid XGB-BAG model exhibited superior accuracy (test R² = 0.901) in comparison to other models. All other models also demonstrate adequate performance (R² greater than 0.80) for the use of predicting the compressive strength of NP-concrete. In addition, the SHapley Additive Explanations (SHAP) study indicated that cement, NPs, and superplasticizers had a beneficial impact on strength. In summary, the research indicates that the hybrid XGB-ADB model, when combined with the indicated input parameters, can effectively forecast the compressive strength of NP-concrete.

Accurate prediction of damage to cables and cabinets in switchgear rooms during fire emergencies is crucial for personal safety and maintenance planning of critical infrastructure. Traditional software tools, such as consolidated fire and smoke transport (CFAST), are used to simulate fire emergencies based on initial fire conditions. However, applying such physics-based simulators to model sequential fire propagation damage to multiple objects is challenging because of significant computational costs. We propose to use machine learning and deep learning-based surrogate models for real-time fire damage prediction of multiple objects in switchgear rooms. To train the surrogate models, we generated a motor cabinet damage dataset using CFAST simulations of fire propagation in the switchgear room. We also use an explainable artificial intelligence (XAI) technique, Shapley additive explanations (SHAP), to interpret relationships between cabinet damage and initial fire conditions. Experimental results demonstrate that the multi-layer perceptron (MLP) is the most effective model for predicting fire damage of multiple objects. Furthermore, SHAP-based interpretations align closely with the sensitivity analysis results, providing reliable and transparent insights into the model's reasoning process. These results highlight our framework provides more robust and efficient results than the existing physics-based simulators, contributing to advancements in fire safety and resilience technology.