Earticle

홈페이지를 통하여 제공되는 인사말에는 보통 추구하고자 하는 목표 등을 포함한 핵심 정보가 담기며 소 통의 일환으로 이용되고 있다. 개인의 건강증진 및 사회통합의 역할을 담당하는 프로스포츠에서도 홈페이지 인사 말은 그 의미하는 바와 역할이 중요하다. 본 연구에서는 국내 프로스포츠 홈페이지 인사말을 대상으로 machine learning 기법을 이용하여 분석해 보았다. 그 결과 다음을 알 수 있었다, 첫째, 분석 대상으로 삼은 4개 프로리그 모두에서 유사한 결과를 보인바, 차별화를 발견할 수 없었다. 둘째, 워드 클라우드 분석, 네트워크 분석을 통해 보 았을 때, 감사의 인사와 함께 지속적인 성원을 부탁하는 내용이 주를 이루고 있는 것으로 나타났다. 나아가 감성 분석 결과 4개 프로리그 모두 예상했던 대로 긍정적인 단어의 선택이 많았으며, 분산분석 결과 그 차이는 통계적 으로 유의하지 않았다(p>0.05). 이에 향후 홈페이지 인사말 개편 시에는, 각 프로리그의 캐치프레이즈를 충분하게 담아내고, 나아가 각 구단이 프로스포츠 운영을 통하여 이루고자 하는 방향성 등이 잘 포함되도록 개선할 것을 제 안하였다.

Greetings on websites homepages typically contain core information, such as goals and visions, and serve as a means of communication. In professional sports, which play a role in promoting individual health and social integration, website greetings hold significant meaning and importance. This study analyzed greetings on domestic professional sports websites using machine learning techniques. The results are as follows: First, all four professional leagues analyzed showed similar results, revealing no differentiation. Second, word cloud and network analyses indicated that greetings mostly consist of words of gratitude and requests for continued support. Furthermore, sentiment analysis showed that, as expected, positive words were predominantly chosen across all four leagues, and analysis of variance revealed no statistically significant differences (p>0.05). Therefore, it is suggested that in future revisions of websites greetings, they should fully incorporate each professional league’s slogan and direction, as well as the objectives each club aims to achieve through professional sports operations.

The paper attempts to document the application of relevant Machine Learning (ML) models on Yelp (a crowd-sourced local business review and social networking site) dataset to analyze, predict and recommend business. Strategically using two cloud platforms to minimize the effort and time required for this project. Seven machine learning algorithms in Azure ML of which four algorithms are implemented in Databricks Spark ML. The analyzed Yelp business dataset contained 70 business attributes for more than 350,000 registered business. Additionally, review tips and likes from 500,000 users have been processed for the project. A Recommendation Model is built to provide Yelp users with recommendations for business categories based on their previous business ratings, as well as the business ratings of other users. Classification Model is implemented to predict the popularity of the business as defining the popular business to have stars greater than 3 and unpopular business to have stars less than 3. Text Analysis model is developed by comparing two algorithms, uni-gram feature extraction and n-feature extraction in Azure ML studio and logistic regression model in Spark. Comparative conclusions have been made related to efficiency of Spark ML and Azure ML for these models.

This study proposes a machine learning approach to understand how post-earnings-announcement drift (PEAD) works. We analyze when PEAD, combined with other factors, becomes more pronounced. To accommodate diverse variables and more complex specifications, two tree-based machine learning approaches including eXtreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LightGBM) are used to examine the relationship between PEAD and 89 variables. The long-short portfolio produced by LightGBM model reports 2.1 times higher returns than the portfolio’s returns, based on the conventional measure of earnings surprise. The model enhances the economic and statistical significance of the long-short portfolio returns. SHapley Additive exPlanations (SHAP) analysis determines feature importance and shows that liquidity, firm size, profitability ratios, share turnover, net trading flows by retail investors, and earnings surprises, play an important role in the prediction of PEAD.

당뇨병은 혈당 수치가 오랜 시간 지속되는 대사질환이다. 당뇨병은 자체 질환에 대한 위험성과 더불어 심혈관과 뇌혈 관, 망막병증, 족부병변 등 질환 다양한 합병증을 유발하는 질병이다. 이렇게 발생하는 합병증으로 인해 사망률을 높이는 원인이 되고 있다. 이에 따라 당뇨병 조기발견과 합병증에 대한 치료가 수반되어야하며, 이를 위해 빠른 진단이 필요하 다. 이에 본 연구에서는 당뇨병 환자 데이터를 이용하여 인공지능 기법 중 로지스틱 회귀분석 머신러닝 알고리즘을 이용한 당뇨병 분류에 대한 머신러닝 모델을 구축하고자 한다. 머신러닝 모델은 scikit-learn을 이용하여 학습용 데이터 80%, 검증용 데이터 20%로 구분하고 교차검증을 5로 나누어 모델을 구성하였다. 로지스틱 회귀분석을 이용한 머신러닝 결과 정확도는 평균 92.24%, AUC 0.95, Recall 0.9294, Precision 0.927, F1 score 0.9248, Kappa 0.6983 이었으 며, 허리-엉덩이 둘레비율이 당뇨병 분류에 가장 큰 영향을 주는 변수였다. 간단한 모델을 통해 높은 성능의 모델을 구축할 수 있었으며, 개인 생활습관 등에 관한 변수를 추가한다면, 더욱 우수한 성능의 머신러닝 모델을 구축할 수 있을 것으로 사료된다.

Diabetes is a metabolic disease in which blood sugar levels last for a long time. Diabetes is a disease that causes various complications of diseases such as cardiovascular and cerebrovascular, retinopathy, and diabetes foot ulcer in addition to the risk of its own disease. Complications that occur in this way are responsible for increasing the mortality rate. Accordingly, early detection of diabetes and treatment for complications should be accompanied, and for this, rapid diagnosis is required. In this study intends to build a machine learning model for diabetes classification using logistic regression machine learning algorithm among artificial intelligence techniques using diabetes patient data. Machine learning model was divided into 80% of training data and 20% of verification data using scikit-learn, and the cross-validation was divided into 5 to construct a model. Accuracy of machine learning results using logistic regression analysis was 92.24% on average, AUC 0.95, Recall 0.9294, Precision 0.927, F1 score 0.9248, and Kappa 0.6983, and the waist-hip circumference ratio was the most influential variable for diabetes classification. A high-performance model could be built through a simple model, and if variables related to personal lifestyle are added, it is believed that a better performance machine learning model can be built.

최근에는 기계 학습, 특히 심층 학습에 많은 연구가 진행되고 있다. 구글, 페이스 북과 같은 대기업이 인공 지능과 기계 학습에 관심을 가지고 있기 때문에, 이러한 연구는 날마다 발전하고있다. 기계 학습은 의학, 번역 및 IT와 같은 다양한 산업에서 사용될 것으로 기대됩니다. 게임 부문은 기계학습 기술적용의 효과가 예상되는 영역 중 하나라고 간주됩니다. 본 논문에서는 MMORPG-Tera의 게임 콘텐츠에서 몬스터의 승패를 예측하는 신경망을 Tensorflow를 통해 설계하였다. 이 모델은 1 개의 입력 레이어, 2 개의 숨겨진 레이어 및 1 개의 출력 레이어를 가지고 있다. 입력 레이어에는 8 개의 노드가 있고 각 숨겨진 레이어에는 16 개의 노드가 있으며 출력 레이어에는 1 개의 노드가 있다. 더 나은 결과를 위해 우리는 그라디언트 디센트, 시그 모이 드 (Sigmoid) 함수 및 Relu 함수 (Activate 함수)에 Adam을 사용한다. 준비된 데이터 세트의 마지막 부분은 테스트 데이터 용으로 사용되고 나머지는 학습 모델 용으로 사용되었다. 이 모델은 5 ~ 10 % 오차 이내의 확률을 예측할 수 있다. 데이터 세트의 부족은 만족스럽지 않은 점으로 남아 있으며, 충분한 데이터가 수집되고 더 개선 된 모델이 준비되면 오류를 더 줄일 수 있다. 그리고 제안된 모델은 앞으로 다른 게임이나 스포츠 게임에도 적용될 것이다.

Recently, much attention has been paid to machine learning - especially deep learning. As big companies like Google, Facebook are interested in AI and machine learning, these research is being developed day by day. Machine learning is expected to be used in various industries such as medical, translation, and IT. The game sector is also considered one of the areas where the effects of applying machine learning technology are expected. In this paper, we designed a Neural Network that predicts the win / loss of monsters in MMORPG-Tera’s Game contents. We designed the model through Tensorflow. This model has 1 input layer, 2 hidden layer and 1 output layer. There are 8 nodes in input layer, 16 nodes in each hidden layer and 1 nodes in output layer. For the better results we use Adam for gradient descent, Sigmoid function and Relu function for Activate function. The last part of the prepared dataset was used for the test data and the rest was used for the learning model. This model is able to predict the odds within 5 ~ 10 % error. The lack of datasets is left as an unsatisfactory point, it will be possible to reduce the error further if sufficient data is acquired and more improved model is prepared. And this proposed model will be applied to other games or sports games in the future.

간세포암 환자에서 TACE 치료 반응은 환자의 개별적인 특성에 따라 달라지므로, 치료 효과를 정확히 예측하는 것은 여전히 중요한 도전 과제로 남아 있다. 이에 본 연구에서는 다양한 머신러닝 알고리즘(Logistic Regression, Linear Support Vector Machine, Random Forest, XGBoost)을 활용하여 TACE 치료 반응을 예측하는 모델을 개발하고, 이들의 성능을 비교하여 가장 우수한 모델을 도출하고자 하였다. 나이, 성별, BMI, 간염 바이러스, 간경변증, 문맥 고혈 압, 생체표지자, CBCT, DAP을 독립변수로, 치료 완전반응을 종속변수로 설정하여 예측 모델을 개발하였다. 성능 평가 결과, 선형 모델 중에서는 Logistic Regression이 가장 높은 성능을 보였으며, 비선형 모델에서는 Random Forest가 재현율(82.12%)에서 높은 값을 나타냈다. 그러나 XGBoost는 정확도(78.57%), 정밀도(81.43%), F1 점수(81.71%)에서 더 높은 성능을 보여 종합적으로 가장 우수한 예측력을 보였다. 이러한 결과는 XGBoost 모델이 복잡한 임상 데이터를 효과적으로 처리할 수 있으며, 향후 TACE 치료 계획 수립에 유용하게 활용될 것으로 사료된다.

Treatment response to TACE (Transarterial chemoembolization) in patients with HCC (Hepatocellular carcinoma) varies depending on individual patient variables, making accurate prediction of treatment outcomes a significant challenge. Therefore, this study aimed to develop Prediction models for treatment response using various machine learning algorithms, including Logistic Regression, Linear Support Vector Machine, Random Forest, and XGBoost, and to compare their performance to identify the most effective model. The Prediction models were developed using independent variables such as age, sex, BMI (Body Mass Index), hepatitis virus, liver cirrhosis, portal hypertension, biomarker, CBCT (Cone Beam Computed Tomography), and DAP (Dose Area Product), with complete treatment response set as the dependent variable. Performance evaluation showed that Logistic Regression had the highest performance among linear models, while Random Forest demonstrated superior recall (82.12%) among nonlinear models. However, XGBoost outperformed other models in terms of accuracy (78.57%), precision (81.43%), and F1 score (81.71%), demonstrating the best overall predictive performance. These results suggest that the XGBoost model can effectively handle complex clinical data and may be useful in supporting future TACE treatment planning.

This study examines the predictability of various machine learning and deep learning models in corporate default forecasts. Using a sample of U.S. corporate defaults over the period of 1963-2020, we find Ensemble classifier and Bi-LSTM classifier forecast the corporate bankruptcy better than other models and the predictability of the Ensemble classifier is more stable in year-to-year variability. Further, machine learning models outperform deep learning models in high yield grade samples, while deep learning models performs better than machine learning models in investment grade samples.

The postal service sector uses machine learning to forecast delivery time and customer traffic. Studies on postal logistics forecasting have used various machine learning algorithms, but there were no attempts using Seasonal and Trend Decomposition using Loess (STL) decomposition, which is frequently used in other fields of time series forecasting. Therefore, this paper proposes a method of applying optimal STL decomposition cycles using the machine learning models of prior studies and the latest machine learning models. First, the proposed method decomposes the daily traffic using STL decomposition to generate three variables (Trend, Seasonal, and Residual). These variables are added to the existing input data variable to train the machine learning model. Finally, a suitable STL decomposition cycle for the model is selected to derive an optimal model. The proposed method was validated by creating nine machine learning (AdaBoost Regression, Random Forest Regression, Ridge, etc.) and two deep learning (DNN, LSTM) models and testing them. As a result, the application of STL decomposition reduced the forecast errors in all models except LSTM. In terms of the proposed method, linear regression had the lowest forecast error, and LSTM had the highest.

The global population is increasing rapidly because of increasing urbanization and such increasing urbanization directs the up-growing need of urban safety and preventions. This urbanization is also responsible for two things that is increased job opportunities and increased the crime rates. In this era technology has gone far more forward in a positive way. By making use of these technologies such as machine learning, artificial intelligence and big data we presented an approach through which crime pattern analysis is done. We have used apache spark (scala-programming) and machine learning algorithm for predictive crime pattern analysis. The data that we have used is a real-world data set based on Chicago city of United State of America. Our main goal of work is to define a predictive crime analysis which shows top crime patterns related to the top community areas of Chicago city.

Employee placement is one of the vital functions of HR, which aligns the employee's skill with the organization's requirements. Traditional methods of placement have many shortcomings: skill mismatching, bias, and underutilization or misutilization of resources. The study uses machine learning (ML) to these problems by evaluating three algorithms-AdaBoost, support vector machine (SVM), and CatBoost-with demographic and job-related data from Kaggle. Results indicate that the maximum accuracy AdaBoost reached was 86%, then SVM with 81.4%, followed by CatBoost at 79%. These findings point to the reliability of AdaBoost in structured data and emphasize the potential that ML has for improving HR efficiency, employee satisfaction, and retention.

In this paper, aiming at the learning deficiency in the given bargaining systems, the decimal code, instead of binary code, is adopted to prevent variables from going beyond limitative scope, which will cause exceptional strategy. Furthermore, a dynamic bargaining system is presented based on machine learning (MLDBS). The result of experiment shows that the agent in MLDBS not can only identify its opponents successfully but can change its strategy in term of different opponents in a bargaining process. It is shown by the experimental datum that MLDBS increase successful times of bargaining and enhance the average payoff of agent.

Choi, J.M.(2023). A gender identification of Korean blog writers through machine learning. Gender identification of texts is a subfield of author analysis; author profiling. This study is an preliminary experiment on an automatic gender detection model for the 1,162 posts of 13 blog owners. As linguistic features, four types of n-gram (word, function word, character, and POS), phoneme frequency, and four lexical sets were chosen, and the support vector machine was adopted as a classifier. The classification accuracy ranged from 54% to 99% depending on the feature type. But the best performing model was produced(obtained) when all the features were inputted combined minus word n-grams. The most salient features distinguishing female from male writers were found to be the first person pronouns( (‘나(I, me)’ and ‘내(+*)’ for females vs. 저(-*)’ and 제(-)’ for males)) and sentence endings(‘다, ‘ᄂ다’ and ‘었다’ for females vs. , ‘습니다’, ‘ᄇ니다’, ‘습니다’, ‘네요’for males). This preliminary study could lead to further research into the gender language variations, and contribute to the development of a stable and robust author profiling system.

This study explores the spatiotemporal expansion patterns of Airbnb hosts by applying the K-shape time-series clustering algorithm to host-level panel data. The analysis covers two periods—pre-pandemic (June 2014–December 2019) and post-pandemic (March 2021–July 2024)—while excluding the COVID-19 disruption phase. Hosts’ property expansion trajectories were examined over 24- and 36-month windows to identify recurring temporal patterns. The results reveal multiple forms of expansion and contraction behaviors, ranging from gradual and sustained growth to temporary decline and recovery. Comparing the two periods shows that post-pandemic host operations became more stable and less volatile. The study contributes to the literature on business expansion and professionalization in short-term rentals and provides practical insights for policymakers and platform managers aiming to foster sustainable and balanced market development.

본 연구에서는 머신러닝 알고리즘을 적용한 모델 별 급성관상동맥증후군 진단 정확도를 비교 평가하고자 한다. 급성 관상동맥증후군 총 857명의 데이터를 학습데이터와 검증데이터로 7 : 3 비율로 구분하고 10 fold로 설정하여 학습에 진행하였다. 학습에 사용된 모델은 random forest, gradient boosting, ada boosting로 학습 후 정확도, AUC, recall, precision, F1 score, kappa로 평가하고, 최종 생성된 모델을 통해 AUC-ROC 곡선과 혼동행렬, 정확도를 비교 평가 하였다. 실험 결과 random forest 모델의 성능이 가장 우수 하였으며, AUC-ROC곡선과 혼동행렬 검증 결과 STEMI 예측성능은 gradient boosting, NSTEMI와 불안정성협심증은 random forest가 가장 예측을 잘 하였다. 검증데이터를 이용한 정확도는 random forest가 100%로 가장 우수한 결과를 도출하였다. 머신러닝 알고리즘을 적용 한 모델에 따른 급성관상동맥증후군 예측은 random forest가 가장 우수하였다. 본 연구결과를 통해 머신러닝 알고리즘 의 보건의료분야 적용에 대한 기초 및 근거 자료로 활용 될 수 있을 것으로 사료되며, 추후 앙상블 기법을 이용한 추가 연구를 통해 효율성을 높일 수 있을 것으로 사료된다.

In this study, we aimed to compare and evaluate the diagnostic accuracy of acute coronary syndrome for each model to which machine learning algorithm was applied. The data of a total of 857 patients with acute coronary syndrome were divided into train dataset and test dataset at a ratio of 7 : 3, and learning was performed by setting 10 folds. The model used for learning is evaluated by accuracy, AUC, recall, precision, F1 score, and kappa after training with random forest, gradient boosting, and ada boosting, and the AUC-ROC curve, confusion matrix, and accuracy are compared through the final model generated. evaluated. As a result of the experiment, the performance of the random forest model was the best, and as a result of the AUC-ROC curve and confusion matrix verification, the prediction performance of STEMI was gradient boosting, and the random forest predicted NSTEMI and unstable angina the best. As for the accuracy using verification data, the random forest produced the best results with 100%. The prediction of acute coronary syndrome according to the model applying the machine learning algorithm was the best in the random forest. It is believed that the results of this study can be used as basic and evidence data for the application of machine learning algorithms to the health care field, and further research using ensemble techniques can improve efficiency.

The purpose of this study was to estimate the forest biomass using satellite imagery and machine learning techniques. In this study, Random Forest, XGBoost, SVM, Multiple Linear Regression were used for forest biomass estimation. Research Forest management plan(8th) data and Sentinel-2 imagery information were used to analysis. As the dependent variables, forest biomass was calculated using volume information, and the biomass expansion factor. The 10 bands of Sentinel-2 were used as independent variable. The optimal forest biomass estimation model was selected by comparing the calculated value based on the Research Forest management plan data and the estimate based on the machine learning techniques. MAE, RMSE, and R2 were calculated for comparison of estimated biomass statistics. As a result, the XGBoost model showed the highest RMSE(61.63ton/ha), MAE(44.16ton/ha), and the highest R2(0.48) value, and was evaluated as the optimal biomass estimation model. The average amount of biomass for sub compartments estimated using the XGBoost model was 225.3tons/ha, which was underestimated by 3.4 tons/ha compared to the average amount of biomass calculated using the Research Forest management plan.

While the gut microbiota is increasingly implicated in the pathogenesis of Parkinson's Disease (PD), the majority of existing research predominantly focuses on Western populations, with limited studies addressing Eastern cohorts. This study aims to elucidate the differential composition of gut microbiota between Eastern and Western PD patients by utilizing advanced machine learning techniques. 16S ribosomal RiboNucleic Acid (rRNA) sequencing data from stool samples are obtained from the Sequence Read Archive (SRA), comprising a random selection of 70 individuals (35 healthy controls and 35 PD patients) from four countries: Korea, Japan, the United States, and Italy. Recursive Feature Elimination (RFE) is employed for feature selection, and four machine learning models— Support Vector Machine (SVM), Random Forest (RF), k- Nearest Neighbors (KNN), and eXtreme Gradient Boosting (XGBoost)—are applied to classify PD patients by geographic origin. RFE identifies 15 key microbial taxa that distinguish between healthy controls and PD patients. Among the models trained on these taxa, the RF model exhibits the highest predictive accuracy, achieving 0.83 ± 0.048. Despite the relatively small sample size, this study underscores the necessity for larger-scale investigations and contributes to a more comprehensive understanding of gut microbiota disparities between Eastern and Western populations in the context of PD.

본 연구는 산업 유사성과 지리적 근접성이 동물병원 지출을 예측하는데 어떤 역할을 하는지 검증하였다. 구체 적으로, 저자들은 머신러닝을 사용하여 산업 유사성이 지리적 근접성보다 더 중요한 역할을 한다는 것을 발견했 다. 특성 중요도를 비교하면 산업 유사성이 지리적 근접성보다 약 5배 정도 더 중요한 것으로 나타났다. 더불어, 본 연구는 지출자의 인구 특성 정보와 거주자의 인구 특성 정보의 중요성도 비교하였다. 그 결과, 일반적으로 지출 자의 인구 특성 정보가 거주자의 인구 특성 정보보다 더 중요하다는 것도 확인하였다. 한편, 거시적인 관점에서 거주자의 인구 특성 정보도 중요성을 지니는 것으로 나타났다.

The authors investigate the role of industrial similarity and geographic proximity in predicting spending on veterinary care. Using machine learning, they find that industrial similarity plays a more substantial role than geographic proximity. Specifically, comparing feature importance reveals that industrial similarity is roughly five times as important as geographic proximity. Further, this study also investigates the importance of spender demographic information and resident demographic information in prediction performance. The result shows that spender demographic information is more important than resident demographic information in general. However, from a macro perspective, resident demographic information also holds importance.

Understanding an accurate trip demand by purpose is crucial for short-term regional planning but also long-term regional planning. In traditional approach, information of purpose-oriented trip demand has been derived from public survey in South Korea such as Travel Diary Survey. This type of data acquiring method maybe useful in a sense that it can capture a meaningful sample regarding to entire country, meanwhile it costs a tremendous amount of budget and time. In this research, we want to offer a novel framework for estimating purposeoriented trip demand with dynamic and effective fashions using data fusion in conjunction with Machine learning techniques. With primary results of this concept, this study showed how several state-of-the-art algorithms, including Deep neural network, UMAP, and random forest in conjunction with Genetic algorithm and Tabu-search for optimization, can contribute to this framework. Although tangible results are yet to come, we expect this framework can contribute to resilience planning such as COVID-19.

본 연구는 의료 데이터 분석에서 단순모형과 복잡모형의 방법론적 우수성을 비교하였다. 두 모형의 우수성 비교를 위한 실증분석을 위해 피 마 인디언 당뇨(Pima Indians Diabetes) 자료를 사용해 0값을 결측으로 처리하고 중앙값 대치·표준화를 거친 뒤 70:30 분할 및 동일 전처리 파 이프라인으로 로지스틱 회귀(단순), 랜덤포레스트, 다층퍼셉트론(MLP)을 학습·평가하였다. 두 모형의 성능평가를 위해서는 임계값 의존 지표 (정확도, 민감도, 특이도, 정밀도(PPV), 음성예측도, 조화평균(F1), 균형정확도, 매튜스 상관계수(MCC))와 임계값 무관·보정 지표(ROC-AUC, PR-AUC, Brier(↓), 보정 절편/기울기, Hosmer-Lemeshow(HL), 임상 순이득(DCA))를 적용하였다. 실증분석 결과, 로지스틱 회귀는 정확도 0.7662, 민감도 0.7160, F1 0.6824, 균형정확도 0.7547, MCC 0.4995로 전반적으로 가장 견조했으며 ROC-AUC 역시 0.8365로 최고였다. 반면 랜덤포레스트는 특이도 0.8733과 PPV 0.6885로 확증 성능이 우수했고, 보정 품질에서도 Brier 0.1594, 보정 절편 0.1356, 기울기 0.9914, HL p=0.7606으로 이상적 수준에 근접했다. 다층퍼셉트론은 동일 조건에서 상대적으로 열세를 보였다. 한편 DCA 차원에서는 전 구간 공통으로 모형 간 격차는 크지 않되, 임계값 선택에 따라 승자가 바뀌었다(0.10: 랜덤포레스트, 0.20~0.30: 로지스틱 회귀). 따라서 실제 적용 시 목표 pt 와 FN/FP 비용구조를 먼저 정하고, 그 범위에서 순이득이 가장 큰 모형-컷오프 조합을 선택하는 것이 합리적이다. 종합하자면, 특정 모형의 방 법론적 우수성은 절대적이지 않고 상대적이며, 분석목적에 따라 선택이 필요하다. 구체적으로 선별(미탐 최소화)이 목표일 때는 로지스틱 회 귀가 적합하며, 확증(위양성 최소화)이나 확률 기반 의사결정에는 보정이 우수한 랜덤포레스트가 적합하다. 따라서 분석모형의 최적 선택은 유병률·FN/FP 비용·임계확률을 반영한 DCA와 함께 이뤄져야 한다.

This study compared the methodological superiority of simple models and complex models in medical data analysis. For an empirical analysis to compare the superiority of the two model classes, we used the Pima Indians Diabetes dataset, treated zeros as missing, performed median imputation and standardization, split the data 70:30, and trained/evaluated logistic regression (simple), random forest, and multilayer perceptron (MLP) under an identical preprocessing pipeline. For performance evaluation of the two model classes, we applied threshold-dependent metrics (accuracy, sensitivity, precision, specificity, F1, balanced accuracy, MCC) and threshold-independent and calibration metrics (ROC-AUC, PR-AUC, Brier, calibration intercept/slope, Hosmer–Lemeshow, clinical net benefit (DCA)). In the empirical results, logistic regression was overall the most robust, with accuracy 0.7662, sensitivity 0.7160, F1 0.6824, balanced accuracy 0.7547, and MCC 0.4995, and also achieved the highest ROC-AUC of 0.8365. By contrast, random forest showed superior rule-in performance with specificity 0.8733 and PPV 0.6885, and was close to an ideal level in calibration quality, with Brier 0.1594, calibration intercept 0.1356, slope 0.9914, and HL p=0.7606. The MLP was relatively inferior under the same conditions. Meanwhile, in the DCA dimension, the gaps between models were not large across the range, but the winner changed with the choice of threshold (0.10: random forest; 0.20–0.30: logistic regression). Therefore, in practical application, it is reasonable to first specify the target pt and the FN/FP cost structure, and then choose the model–cutoff combination that yields the largest net benefit within that range. In sum, the methodological superiority of a given model is not absolute but relative, and selection should depend on the analytic objective. Specifically, when the goal is screening (minimizing missed positives), logistic regression is appropriate, whereas for rule-in (minimizing false positives) or probability-based decision making, random forest with superior calibration is suitable. The optimal choice of analysis model should be made together with DCA that reflects prevalence, FN/FP costs, and the threshold probability.