• Korean Chemical Engineering Research
• /
• 제61권4호
• /
• pp.542-549
• /
• 2023

대안 모델링에 대한 관심이 커진 이후 데이터 기반의 기계학습을 이용하여 비선형 화학 공정을 모사하고자 하는 연구가 지속되고 있다. 그러나 기계 학습 모델의 black box 성질로 인하여 모델의 해석 가능성에 한계는 산업 적용에 걸림돌이 되고 있다. 따라서, 모델의 정확도가 보장된 상태에서 해석력을 부여하는 개념인 설명 가능한 인공지능(explainable artificial intelligence, XAI)을 이용하여 화학 공정 분석을 시도하고자 한다. 기존의 화학 공정 민감도 분석이 변수의 민감도 지수를 계산하고 순위를 매기는 데에 그쳤다면, XAI를 이용하여 전역적, 국소적 민감도 분석뿐만 아니라 변수들 간의 상호작용에 대하여 분석하여 데이터로부터 물리적 통찰을 얻어내는 방법론을 제안한다. 사례 연구의 대상공정인 암모니아 합성 공정에 대하여 첫번째 반응기로 향하는 흐름에 대한 예열기(preheater)의 온도, 세 반응기로 향하는 cold-shot의 분배 비율을 공정 변수로 설정하였다. Matlab과 Aspen plus를 연동하여 공정 변수를 바꿔가면서 암모니아의 생산량과 세 반응기의 최고 온도에 대한 데이터를 얻었으며, tree 기반의 모델들을 훈련시켰다. 그리고 성능이 좋은 모델에 대하여 XAI 기법 중 하나인 SHAP 기법을 이용하여 민감도 분석을 수행하였다. 전역적 민감도 분석 결과, 예열기의 온도가 가장 큰 영향을 미쳤으며 국소적 민감도 분석 결과에서 생산성 향상 및 과열 방지를 위한 공정 변수들의 범위를 규정할 수 있었다. 이처럼 화학 공정의 대안 모델을 구축하고 설명 가능한 인공지능을 이용해 민감도 분석을 진행하는 방법론을 통해 공정 최적화에 대한 정량적, 정성적 피드백을 제안하는 데 도움을 줄 것이다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제17권8호
• /
• pp.2016-2029
• /
• 2023

Accurate prediction of critical illness is significant for ensuring the lives and health of patients. The selection of indicators affects the real-time capability and accuracy of the prediction for critical illness. However, the diversity and complexity of these indicators make it difficult to find potential connections between them and critical illnesses. For the first time, this study proposes an indicator analysis model to extract key indicators from the preoperative and intraoperative clinical indicators and laboratory results of critical illnesses. In this study, preoperative and intraoperative data of heart failure and respiratory failure are used to verify the model. The proposed model processes the datum and extracts key indicators through four parts. To test the effectiveness of the proposed model, the key indicators are used to predict the two critical illnesses. The classifiers used in the prediction are light gradient boosting machine (LightGBM) and eXtreme Gradient Boosting (XGBoost). The predictive performance using key indicators is better than that using all indicators. In the prediction of heart failure, LightGBM and XGBoost have sensitivities of 0.889 and 0.892, and specificities of 0.939 and 0.937, respectively. For respiratory failure, LightGBM and XGBoost have sensitivities of 0.709 and 0.689, and specificity of 0.936 and 0.940, respectively. The proposed model can effectively analyze the correlation between indicators and postoperative critical illness. The analytical results make it possible to find the key indicators for postoperative critical illnesses. This model is meaningful to assist doctors in extracting key indicators in time and improving the reliability and efficiency of prediction.

• Geomechanics and Engineering
• /
• 제38권5호
• /
• pp.517-528
• /
• 2024

For numerous tunnelling projects implemented in urban areas due to limited space, it is crucial to take into account the interaction between the foundation, ground, and tunnel. In predicting the deformation of piled foundations and the ground during twin tunnel excavation, it is essential to consider various factors. Therefore, this study derived a prediction model for pile group settlement using machine learning to analyze the importance of various factors that determine the settlement of piled foundations during twin tunnelling. Laboratory model tests and numerical analysis were utilized as input data for machine learning. The influence of each independent variable on the prediction model was analyzed. Machine learning techniques such as data preprocessing, feature engineering, and hyperparameter tuning were used to improve the performance of the prediction model. Machine learning models, employing Random Forest (RF), eXtreme Gradient Boosting (XGB), and Light Gradient Boosting Machine (LightGBM, LGB) algorithms, demonstrate enhanced performance after hyperparameter tuning, particularly with LGB achieving an R² of 0.9782 and RMSE value of 0.0314. The feature importance in the prediction models was analyzed and P_N was the highest at 65.04% for RF, 64.81% for XGB, and P_CTC (distance between the center of piles) was the highest at 31.32% for LGB. SHAP was utilized for analyzing the impact of each variable. P_N (the number of piles) consistently exerted the most influence on the prediction of pile group settlement across all models. The results from both laboratory model tests and numerical analysis revealed a reduction in ground displacement with varying pillar spacing in twin tunnels. However, upon further investigation through machine learning with additional variables, it was found that the number of piles has the most significant impact on ground displacement. Nevertheless, as this study is based on laboratory model testing, further research considering real field conditions is necessary. This study contributes to a better understanding of the complex interactions inherent in twin tunnelling projects and provides a reliable tool for predicting pile group settlement in such scenarios.

• Steel and Composite Structures
• /
• 제51권6호
• /
• pp.679-695
• /
• 2024

This paper presents a novel technique that combines machine learning (ML) with moth-flame optimization (MFO) methods to predict the axial compressive strength (ACS) of concrete filled double skin steel tubes (CFDST) columns. The proposed model is trained and tested with a dataset containing 125 tests of the CFDST column subjected to compressive loading. Five ML models, including extreme gradient boosting (XGBoost), gradient tree boosting (GBT), categorical gradient boosting (CAT), support vector machines (SVM), and decision tree (DT) algorithms, are utilized in this work. The MFO algorithm is applied to find optimal hyperparameters of these ML models and to determine the most effective model in predicting the ACS of CFDST columns. Predictive results given by some performance metrics reveal that the MFO-CAT model provides superior accuracy compared to other considered models. The accuracy of the MFO-CAT model is validated by comparing its predictive results with existing design codes and formulae. Moreover, the significance and contribution of each feature in the dataset are examined by employing the SHapley Additive exPlanations (SHAP) method. A comprehensive uncertainty quantification on probabilistic characteristics of the ACS of CFDST columns is conducted for the first time to examine the models' responses to variations of input variables in the stochastic environments. Finally, a web-based application is developed to predict ACS of the CFDST column, enabling rapid practical utilization without requesting any programing or machine learning expertise.

• Journal of Trauma and Injury
• /
• 제37권3호
• /
• pp.201-208
• /
• 2024

Purpose: The number of elderly patients with trauma is increasing; therefore, precise models are necessary to estimate the mortality risk of elderly patients with trauma for informed clinical decision-making. This study aimed to develop machine learning based predictive models that predict 30-day mortality in severely injured elderly patients with trauma and to compare the predictive performance of various machine learning models. Methods: This study targeted patients aged ≥65 years with an Injury Severity Score of ≥15 who visited the regional trauma center at Chungbuk National University Hospital between 2016 and 2022. Four machine learning models-logistic regression, decision tree, random forest, and eXtreme Gradient Boosting (XGBoost)-were developed to predict 30-day mortality. The models' performance was compared using metrics such as area under the receiver operating characteristic curve (AUC), accuracy, precision, recall, specificity, F1 score, as well as Shapley Additive Explanations (SHAP) values and learning curves. Results: The performance evaluation of the machine learning models for predicting mortality in severely injured elderly patients with trauma showed AUC values for logistic regression, decision tree, random forest, and XGBoost of 0.938, 0.863, 0.919, and 0.934, respectively. Among the four models, XGBoost demonstrated superior accuracy, precision, recall, specificity, and F1 score of 0.91, 0.72, 0.86, 0.92, and 0.78, respectively. Analysis of important features of XGBoost using SHAP revealed associations such as a high Glasgow Coma Scale negatively impacting mortality probability, while higher counts of transfused red blood cells were positively correlated with mortality probability. The learning curves indicated increased generalization and robustness as training examples increased. Conclusions: We showed that machine learning models, especially XGBoost, can be used to predict 30-day mortality in severely injured elderly patients with trauma. Prognostic tools utilizing these models are helpful for physicians to evaluate the risk of mortality in elderly patients with severe trauma.

• 품질경영학회지
• /
• 제50권3호
• /
• pp.459-471
• /
• 2022

Purpose: The purpose of this study was to increase prediction accuracy for an anomaly interval identified using an artificial intelligence-based time series anomaly detection technique by establishing a pre-processing process. Methods: Significant variables were extracted by applying feature selection techniques, and anomalies were derived using the TadGAN time series anomaly detection algorithm. After applying machine learning and deep learning methodologies using normal section data (excluding anomaly sections), the explanatory power of the anomaly sections was demonstrated through performance comparison. Results: The results of the machine learning methodology, the performance was the best when SHAP and TadGAN were applied, and the results in the deep learning, the performance was excellent when Chi-square Test and TadGAN were applied. Comparing each performance with the papers applied with a Conventional methodology using the same data, it can be seen that the performance of the MLR was significantly improved to 15%, Random Forest to 24%, XGBoost to 30%, Lasso Regression to 73%, LSTM to 17% and GRU to 19%. Conclusion: Based on the proposed process, when detecting unsupervised learning anomalies of data that are not actually labeled in various fields such as cyber security, financial sector, behavior pattern field, SNS. It is expected to prove the accuracy and explanation of the anomaly detection section and improve the performance of the model.

• International Journal of Computer Science & Network Security
• /
• 제22권12호
• /
• pp.125-131
• /
• 2022

The diversity and complexity of criminal, administrative and civil cases resolved by the courts makes it difficult to develop universal automated tools for the analysis and evaluation of justice. However, big data generated in the scope of justice gives hope that this problem will be resolved as soon as possible. The big data applying makes it possible to identify typical options for resolving cases, form detailed rules for the individualization of a court decision, and correlate these rules with an abstract provisions of law. This approach allows us to somewhat overcome the contradiction between the abstract and the concrete in law, to automate the analysis of justice and to model e-justice for scientific and practical purposes. The article presents the results of using dimension reduction, SHAP value, and p-value to identify, analyze and evaluate the individualization of justice and the differentiation of legal regulation. Processing and analysis of arrays of court decisions by computational methods make it possible to identify the typical views of courts on questions of fact and questions of law. This knowledge, obtained automatically, is promising for the scientific study of justice issues, the improvement of the prescriptions of the law and the probabilistic prediction of a court decision with a known set of facts.

• 한국정보처리학회:학술대회논문집
• /
• 한국정보처리학회 2022년도 추계학술발표대회
• /
• pp.55-56
• /
• 2022

Hypertension is a severe health problem and increases the risk of other health issues, such as heart disease, heart attack, and stroke. In this research, we propose a machine learning-based prediction method for the risk of chronic hypertension. The proposed method consists of four main modules. In the first module, the linear interpolation method fills missing values of the integration of gas and meteorological datasets. In the second module, the OrdinalEncoder-based normalization is followed by the Decision tree algorithm to select important features. The prediction analysis module builds three models based on k-Nearest Neighbors, Decision Tree, and Random Forest to predict hypertension levels. Finally, the features used in the prediction model are explained by the DeepSHAP approach. The proposed method is evaluated by integrating the Korean meteorological agency dataset, natural gas leakage dataset, and Korean National Health and Nutrition Examination Survey dataset. The experimental results showed important global features for the hypertension of the entire population and local components for particular patients. Based on the local explanation results for a randomly selected 65-year-old male, the effect of hypertension increased from 0.694 to 1.249 when age increased by 0.37 and gas loss increased by 0.17. Therefore, it is concluded that gas loss is the cause of high blood pressure.

• 지능정보연구
• /
• 제29권1호
• /
• pp.121-142
• /
• 2023

수입식품의 수입 건수와 수입 중량이 꾸준히 증가함에 따라 식품안전사고 방지를 위한 수입식품의 안전관리가 더욱 중요해지고 있다. 식품의약품안전처는 통관단계의 수입검사와 더불어 통관 전 단계인 해외제조업소에 대한 현지실사를 시행하고 있지만 시간과 비용이 많이 소요되고 한정된 자원 등의 제약으로 데이터 기반의 수입식품 안전관리 방안이 필요한 실정이다. 본 연구에서는 현지실사 전 부적합이 예상되는 업체를 사전에 선별하는 기계학습 예측 모형을 마련하여 현지실사의 효율성을 높이고자 하였다. 이를 위해 통합식품안전정보망에 수집된 총 303,272건의 해외제조가공업소 기본정보와 2019년도부터 2022년 4월까지의 현지실사 점검정보 데이터 1,689건을 수집하였다. 해외제조가공업소의 데이터 전처리 후 해외 제조업소_코드를 활용하여 현지실사 대상 데이터만 추출하였고, 총 1,689건의 데이터와 103개의 변수로 구성되었다. 103개의 변수를 테일유(Theil-U) 지표를 기준으로 '0'인 변수들을 제거하였고, 다중대응분석(Multiple Correspondence Analysis)을 적용해 축소 후 최종적으로 49개의 특성변수를 도출하였다. 서로 다른 8개의 모델을 생성하고, 모델 학습 과정에서는 5겹 교차검증으로 과적합을 방지하고, 하이퍼파라미터를 조정하여 비교 평가하였다. 현지실사 대상업체 선별의 연구목적은 부적합 업체를 부적합이라고 판정하는 확률인 검측률(recall)을 최대화하는 것이다. 머신러닝의 다양한 알고리즘을 적용한 결과 Recall_macro, AUROC, Average PR, F1-score, 균형정확도(Balanced Accuracy)가 가장 높은 랜덤포레스트(Random Forest)모델이 가장 우수한 모형으로 평가되었다. 마지막으로 모델에 의해서 평가된 개별 인스턴스의 부적합 업체 선정 근거를 제시하기 위해 SHAP(Shapley Additive exPlanations)을 적용하고 현지실사 업체 선정 시스템에의 적용 가능성을 제시하였다. 본 연구결과를 바탕으로 데이터에 기반한 과학적 위험관리 모델을 통해 수입식품 관리체계의 구축으로 인력·예산 등 한정된 자원의 효율적 운영방안 마련에 기여하길 기대한다.

• 지능정보연구
• /
• 제28권2호
• /
• pp.147-170
• /
• 2022

특허의 사적 가치(private value)를 나타내는 특허수명 추정은 오래전부터 연구되었으나 추정과정에서 선형모델에 의존하는 경우가 대부분이었고, 기계학습 방법을 사용하더라도 변수 간 관계에 대한 해석이나 설명이 부족하였다. 본 연구에서는 특허의 생존 기간이 특허의 가치를 대리한다는 기존 연구결과를 바탕으로 특허 등록 이후의 생존 기간(연장횟수) 예측을 통해 특허의 가치를 추정한다. 이를 위해 1996~2017년까지 미국 특허청(USPTO)에 출원하여 등록된 특허 4,033,414개를 수집하였다. 특허수명을 예측하기 위해 기존 연구에서 특허수명에 영향을 미친다고 밝혀진 특허의 특성, 특허의 소유자 특성, 특허의 발명가 특성을 반영할 수 있는 다양한 변수가 사용되었다. 서로 다른 4개의 모델(Ridge Regression, Random Forest, Feed-forward Neural Network, Gradient Boosting Models)을 생성하고, 모델 학습 과정에서는 5-fold Cross Validation으로 초매개변수 조정이 이루어졌다. 이후 생성된 모델의 성능을 평가하고 특허수명을 추정할 수 있는 예측변수의 상대적 중요도를 제시하였다. 또한, 성능이 우수했던 Gradient Boosting Model을 기반으로 Accumulated Local Effects Plot을 제시하여 예측변수와 특허수명 간 관계를 시각적으로 나타내었다. 마지막으로 모델에 의해서 평가된 개별 특허의 평가 근거를 제시하기 위하여 Kernal SHAP(SHapley Additive exPlanations)을 적용하고 특허평가 시스템에의 적용 가능성을 제시한다. 본 연구는 기존에 특허수명을 추정하는 연구에 누적적으로 기여한다는 점 그리고 선형성을 바탕으로 진행된 기존 특허수명 추정 연구들의 한계를 보완하고 복잡한 비선형 관계를 설명가능한 방식으로 제시하였다는 점에서 학문적 의의가 있다. 또한, 개별 특허의 평가 근거를 도출하는 방법을 소개하고 특허평가 시스템에의 적용 가능성을 제시하였다는 점에서 실무적 의의가 있다.