Earticle

Support vector machines (SVMs) were originally designed for binary classification problems. However, some real-world multiclass problems including corporate bond rating can’t be solved by the binary classifier. For this reason, many researchers have tried to extend original SVM to multiclass classification. However, their studies have only focused on classifying samples into nominal categories. But, their approaches also have a limitation because there are multiclass classification problems whose classes are ordinal in real world. In this study, we propose ordinal multiclass SVMs which apply ordinal pairwise partitioning (OPP) to conventional SVMs in order to handle ordinal multiple classes efficiently and effectively. Our suggested model may use less classifier but predict more accurately because it utilizes characteristics of the order of the classes. We apply it to the real-world customer classification case for validating the proposed model. The result shows that the proposed multiclass SVM models improve classification performance.

Conventional Support Vector Machines (SVMs) have been utilized as classifiers for binary classification problems. However, certain real world problems, including corporate bond rating, cannot be addressed by binary classifiers because these are multi-class problems. For this reason, numerous studies have attempted to transform the original SVM into a multiclass classifier. These studies, however, have only considered nominal classification problems. Thus, these approaches have been limited by the existence of multiclass classification problems where classes are not nominal but ordinal in real world, such as corporate bond rating and multiclass customer classification. In this study, we adopt a novel multiclass SVM which can address ordinal classification problems using ordinal pairwise partitioning (OPP). The proposed model in our study may use fewer classifiers, but it classifies more accurately because it considers the characteristics of the order of the classes. Although it can be applied to all kinds of ordinal multiclass classification problems, most prior studies have applied it to finance area like bond rating. Thus, this study applies it to a real world customer level classification case for implementing customer relationship management. The result shows that the ordinal multiclass SVM model may also be effective for customer level classification.

지금까지 금융기관의 신용위험관리를 효과적이고 효율적으로 수행하기 위한 지능형 기술에 대한 연구가 활발하게 이루어져 왔다. 특히 기업의 부실 예측이나 신용등급 예측 문제를 해결하는데 있어, 다양한 기계학습 분류 알고리즘들이 제안되고 적용되어 왔는데, 이들은 효과적인 입력변수 선정에 대한 해답을 제공해 주지 못하고, 과적합 및 잡음·이상치에 취약할 수 있는 위험이 있으며, 각종 모수에 대한 조정 작업이 요구되는 한계점이 있었다. 이러한 기존 기법들의 한계를 극복하기 위한 대안으로 본 연구는 랜덤 포레스트(Random Forests)의 응용을 제안한다. 미국의 통계학자 Breiman(2001)에 의해 제안된 랜덤 포레스트는 배깅(bagging)과 무작위 입력 변수 선택(random input selection)을 통해 단일 의사결정 나무를 다수의 의사결정 나무로 확장시킨 기계학습 기법이다. 랜덤 포레스트는 과적합에서 자유로우며, 잡음이나 이상치에 받는 영향이 적고, 높은 정확도를 얻을 수 있는 특징을 지닌다. 본 연구에서는 기업신용등급 예측에 랜덤 포레스트 기법을 적용해 보고, 그 성과를 전통적으로 사용되어 온 다른 기법들과 비교하여 예측성과의 개선이 있는지 확인해 보고자 하였다. 이를 위해 본 연구에서는 1,295개 국내 상장기업을 대상으로 하는 기업신용등급 평가모형 구축에 랜덤 포레스트를 적용해 보았다. 제안된 기법의 성과를 비교평가하기 위한 전통적인 분류 기법으로는 다중판별분석, 인공신경망, 다분류 SVM 모형을 사용하였다. 실증분석 결과, 제안 기법이 전통적인 기법들과 비교해, 보다 정확한 예측결과를 산출함을 확인할 수 있었다.

There has been a lot of research on intelligent information processing techniques to effectively and efficiently manage credit risk in the financial industry. To predict corporate credit ratings better, various algorithms of machine learning have been proposed and applied. However, the algorithms have some limitations. First of all, they are unable to give a solution for which variables in the dataset should be considered as input variables to effectively predict a dependent variable. Also, they are easily exposed to overfitting and risk being affected by outliers and noise. In addition, they are required to tune their numerous parameters. Under this background, this study proposes the application of Random Forests(RF) to overcome the weakness of the traditional techniques. RF were designed by Leo Breiman, an American statistician and professor, and are an ensemble learning technique. This algorithm has both features originating from bagging algorithm, which is to extend a single decision tree to various decision trees, and from random input selection. The benefit of our proposed algorithm is free from an overfitting issue, and both noise and outliers in the dataset cannot have an effect on it. Thus, our model can yield high accuracy in forecasting corporate credit ratings. To validate the effectiveness of RF in predicting corporate credit ratings, we applied it to a real world dataset of corporate credit rating in Korea. The data that we used for this research consisted of various financial ratios of 1,295 corporations and their corporate credit ratings. To evaluate and compare our model with traditional methods, we used the following methods: MDA(Multiple Discriminant Analysis), ANN(Artificial Neural Network) and MSVM(Multiclass SVM). Each model’s performance was calculated by predictive accuracy based on training and validation.　Empirical results showed that RF outperforms other traditional methods from the perspective of prediction accuracy.

Transfer learning significantly improves the performance of a deep learning model on challenging datasets. However, the pre-trained models have certain constraints in terms of their architecture. For example, the state-of-the-art pre-trained models expect an input image with three-color channels because of the wide availability of color images. However, there are certain domains, e.g., medical applications, where grayscale images are produced and the models are required to perform certain tasks on them. Therefore, in this work we propose an approach to run pre-trained models on grayscale images while benefiting from transfer learning for multiclass classification task. We have used the MobileNetV2 pre-trained model to classify the CIFAR datasets. We have compared our results with a conventional method where the grayscale image is stacked up to form a pseudo-color image. Our analysis have shown that the proposed method reduces the computational time per epoch while improves the accuracy of the model.

본 연구는 넷플릭스의 계정 공유 제한 강화 정책에 대한 이용자의 대응 행동을 예측하기 위해 기계학습 기반 다중 분류 모델을 구축하였다. 한국언론진흥재단의 설문 자료(n=681)를 활용하여 이용자 행동을 혼자 이용, 추가 요금 지불, 공유 중단, 편법 모색, 이탈의 5개 범주로 분류하고, 6가지 알고리즘을 비교한 결과 Random Forest가 정확도 0.664, Macro F1-score 0.602로 가장 우수한 성능을 보였으며, 이탈(89%)과 편법 모색(83%) 그룹을 높은 민감도로 탐지하였다. 순열 중요도(PI, Permutation Importance) 및 SHAP 분석 결과, 구독 구조(32.6%)와 공유 동기(25.1%)가 인구통계학적 변수(8.1%)보다 훨씬 큰 설명력을 보였다. 계정 소유자는 이탈 성향이, 공유 수혜자는 편법 모색 성향이 나타났다. 본 연구는 설명 가능한 AI 기법을 통해 이용자 행동을 다차원적으로 예측함으로써 OTT 서비스 의 정책 수립에 실질적 시사점을 제공한다.

This study developed a machine learning-based multi-class classification model to predict user responses to Netflix's account-sharing restriction policy. Using survey data from the Korea Press Foundation (n=681), user behaviors were categorized into five types: individual use, paying additional fees, stopping sharing, workaround-seeking, and service churn. Six algorithms were compared, with Random Forest achieving the highest performance (Accuracy 0.664, Macro F1-score 0.602) and high sensitivity in detecting churn (89%) and workaround-seeking (83%) groups. Permutation Importance and SHAP analyses revealed that subscription structure (32.6%) and sharing motivations (25.1%) had substantially greater explanatory power than demographic variables (8.1%). Account owners exhibited strong churn tendencies, while sharing beneficiaries sought workarounds. This study extends platform policy research by multidimensionally predicting user behaviors through explainable AI techniques, providing practical implications for OTT service policy formulation.

Reliable and fine-grained musical metadata are required for efficient search of rapidly increasing music files. In particular, since the primary motive for listening to music is its emotional effect, diversion, and the memories it awakens, emotion classification along with genre classification of music is crucial. In this paper, as an initial approach towards a “ground-truth” dataset for music emotion and genre classification, we elaborately generated a music corpus through labeling of a large number of ordinary people. In order to verify the suitability of the dataset through the classification results, we extracted features according to MPEG-7 audio standard and applied different machine learning models based on statistics and deep neural network to automatically classify the dataset. By using standard hyperparameter setting, we reached an accuracy of 93% for genre classification and 80% for emotion classification, and believe that our dataset can be used as a meaningful comparative dataset in this research field.

An extremely crucial step in the diagnosis of cancers is to select a small number of informative genes for accurate classification. This issue has become a hot focus in the data mining of gene expression profiles. Especially for data with a large number of cancer types, many conventional classification methods show very poor performance. Here, we proposed a new approach for gene selection and multi-cancer classification based on step-by-step improvement of classification performance (SSiCP). The SSiCP gene selection algorithms were evaluated over the NCI60 and GCM benchmark datasets, with accuracy of 96.6% and 95.5% in 10-fold cross-validation, respectively. Furthermore, the SSiCP outperformed recently published algorithms when applied to another two multi-cancer data sets. Computational evidence indicated that SSiCP can avoid overfitting effectively. Compared with various gene selection algorithms, the implementation of SSiCP is simple and many of the selected genes by SSiCP are shown to be closely related to cancers.

This paper proposes a Multiclass Least Squares Twin Support Vector Machine (MLSTSVM) classifier for multi-class classification problems. The formulation of MLSTSVM is obtained by extending the formulation of recently proposed binary Least Squares Twin Support Vector Machine (LSTSVM) classifier. For M-class classification problem, the proposed classifier seeks M-non parallel hyper-planes, one for each class, by solving M-linear equations. A regularization term is also added to improve the generalization ability. MLSTSVM works well for both linear and non-linear type of datasets. It is relatively simple and fast algorithm as compared to the other existing approaches. The performance of proposed approach has been evaluated on twelve benchmark datasets. The experimental result demonstrates the validity of proposed MLSTSVM classifier as compared to the typical multi-classifiers based on ‘Support Vector Machine’ and ‘Twin Support Vector Machine’. Statistical analysis of the proposed classifier with existing classifiers is also performed by using Friedman’s Test statistic and Nemenyi post hoc techniques.

In this paper we propose a new method for solving multiclass problem with least squares support vector machine(LS-SVM) regression. This method implements one-against-all scheme which is as accurate as any other approach. We also propose cross validation(CV) method to select effectively the optimal values of hyper-parameters which affect the performance of the proposed multiclass method. Experimental results are then presented which indicate the performance of the proposed multiclass method.

Purpose: Our study aims to develop a machine learning model that not only predicts the occurrence of post-induction hypotension (PIH) but also assesses its severity, addressing a broader patient population than previous studies which mostly focused on a single population. Materials and Methods: In our study, we extracted data from 71473 patients aged 18 years and older who underwent general anesthesia for non-cardiac surgery at a tertiary care hospital, using the electronic medical record system for modeling. We used patient demographics, baseline and pre-induction blood pressure, preoperative laboratory tests, surgical details, and anesthetics data, focusing on predicting PIH. The severity of hypotension was assessed by integrating and calculating the integral value of hypotensive periods. We employed several machine learning techniques and evaluated their performance using accuracy, precision, F1-scores, and macro-averaged area under the curve. Additionally, SHapley Additive exPlanation values were used to determine the key factors influencing the predictions. Results: A multiclass classification model, which assesses hypotension severity through the area of hypotension, surpassed the binary model with an F1-score of 0.664. Among various machine learning techniques, the eXtreme Gradient Boosting (XGBoost) model exhibited superior prediction performance, achieving an accuracy of 0.755 and an F1-score of 0.664. Models using preoperative blood pressure and demographic data outperformed those using other datasets. Conclusion: We found that using the XGBoost ensemble machine learning technique aids in predicting PIH before surgery, and introducing a multiclass classification model that indicates the severity of hypotension to clinicians leads to an overall enhancement in prediction performance, thereby increasing its clinical utility for real-world applications.

현재 마이크로어레이 기술은 대량의 유전자 발현 데이터 특히 암과 관련한 데이터들을 쏟아내고 있다. 이 데이터를 기반으로 암의 종류에 따른 유전자들의 차별적 발현 양상을 분석하고 발현량의 변화가 두드러지는 유전자들에 기반하여 암을 분별할 수 있는 분류 모델을 구축한 후, 이것을 암을 진단하거나 예측하는데 이용할 수 있다. 본 논문에서는 마이크로어레이 데이터를 사용해 특징추출방법과 분류를 위한 Naive Bayes, k-Nearest Neighborhood, Decision Tree, Support Vector Machine, Neural Network 알고리즘을 이용하여 최적의 조합을 찾고 어떤 알고리즘이 가장 효과적인지 실험을 통해 분석해보고 성능평가 하는 것을 목표로 한다.

로그 데이터는 정보 시스템의 주요 동작과 상태를 이해하고 판단하는 근거로 사용되어 왔으며, 여러 보안 분야 응용에서도 중요한 입력 데이터로 사용된다. 로그 데이터로부터 필요한 정보를 얻어 이를 근거로 의사 결정을 하고, 적절한 대응 방안을 취하는 것은 시스템을 보호하고 안정적으로 운영하는 데 있어 필수적인 요소이지만, 로그의 종류와 양이 폭발적으로 증가함에 따라 기존 도구들로는 효과적이고 효율적인 대응이 쉽지 않은 상황이다. 이에 본 연구에서는 자연어 처리 기반의 머신 러닝을 이용해 멀티 소스 이벤트 로그의 보안 심각도를 여러 단계로 분류하는 방법을 제안하였으며, 472,972건의 훈련 및 테스트 샘플을 이용하여 실험을 수행한 결과 99.59%의 정확도를 달성하였다.

Log data has been used as a basis in understanding and deciding the main functions and state of information systems. It has also been used as an important input for the various applications in cybersecurity. It is an essential part to get necessary information from log data, to make a decision with the information, and to take a suitable countermeasure according to the information for protecting and operating systems in stability and reliability, but due to the explosive increase of various types and amounts of log, it is quite challenging to effectively and efficiently deal with the problem using existing tools. Therefore, this study has suggested a multiclass classification of the security severity level of multi-source event log using machine learning based on natural language processing. The experimental results with the training and test samples of 472,972 show that our approach has archived the accuracy of 99.59%.

In this study, we present a system for identifying equivalent grades of standardized wire rod steel based on alloy composition using machine learning techniques. The system comprises two models, one based on a supervised multi-class classification algorithm and the other based on unsupervised autoencoder algorithm. Our evaluation showed that the supervised model exhibited superior performance in terms of prediction stability and reliability of prediction results. This system provides a useful tool for non-experts seeking similar grades of steel based on alloy composition.

본 논문에서는 초고차원 자료의 다항분류를 위한 변수선별 방법에 대해 비교 연구를 진행하였다. 다항분류를 위한 변수선별 방법에는 일대일 혹은 일대다 비교를 통해 이항분류를 위한 방법을 확장시켜 적용하는 방법과 다항 반응 변수에 직접 적용할 수 있는 방법이 있다. 다항분류를 위한 변수선별 성능을 확인하기 위하여 여러가지 상황-설명변수의 꼬리가 두꺼운 경우, 신호변수와 잡음변수가 서로 연관된 경우, 결합분포상으로 연관되어 있지만 주변분포 상으로는 연관되어 있지 않은 경우, 다범주 반응변수의 분포가 불균형인 경우-을 가정하고 모의실험을 진행하였고, 실제 자료에도 적용해 보았다. 그 결과, 모형 가정을 필요로 하지 않는 방법들이 안정적인 성능을 보이는 것을 확인하였다.

We compare various variable screening methods on multiclass classification problems when the data is ultrahigh-dimensional. Two different approaches were considered: (1) pairwise extension from binary classification via one versus one or one versus rest comparisons and (2) direct classification of multiclass responses. We conducted extensive simulation studies under different conditions: heavy tailed explanatory variables, correlated signal and noise variables, correlated joint distributions but uncorrelated marginals, and unbalanced response variables. We then analyzed real data to examine the performance of the methods. The results showed that model-free methods perform better for multiclass classification problems as well as binary ones.

다중 클래스 문서분류는 주어진 여러 개의 관심사별로 문서를 선별해 주는 문제이다. 문서 분류 문제의 특징은 문서가 매우 높은 차원으로 표현된다는 것이다. 다른 학습 알고리즘에 비해 SVM 알고리즘은 차원을 전혀 줄이지 않고 문제를 해결한다. 본 논문에서는 SVM 학습 알고리즘을 이용하여 대규모의 뉴스 그룹 문서 분류 문제를 다룬다. 다중 클래스 문서 분류를 위해서 각 클래스에 대한 SVM학습 결과를 효과적으로 결합하였으며 실험을 통하여 SVM과 다른 학습 알고리즘과의 성능을 비교하였다.

본 논문에서는 변환 학습에 기반을 둔 다중 클래스 영상 장면 분류 방법을 제시한다. ImageNet 대형 이미지 데이터 세트에서 사전 훈련된 네트워크 모델에 의존하여 다중 클래스의 자연 장면 이미지를 분류한다. 실험에서는 최적화된 ResNet 모델을 Kaggle의 Intel Image Classification 데이터 셋에 분류하여 우수한 결과를 얻었다.

In this paper, we present a multi-class image scene classification method based on transformation learning. ImageNet classifies multiple classes of natural scene images by relying on pre-trained network models on large image datasets. In the experiment, we obtained excellent results by classifying the optimized ResNet model on Kaggle's Intel Image Classification data set.

본 논문에서는 지도 학습에 기반을 둔 다중 클래스 이미지 장면 분류 방법을 제시한다. 데이터 세트에서 콘볼루션 뉴런 네트워크 모델에 학습시켜 다중 클래스 사람의 표정 장면 이미지를 분류하였으며, 실험에서는 최적화된 CNN 모델을 Google image 데이터 세트에 분류하여 유의미한 결과를 얻을 수 있었다.

In this paper, we present a multi-class image scene classification method based on map learning. We were able to learn from the convolutional neural network model in the dataset, classify facial scene images of multiclass people, and classify the optimized CNN model into the Google image dataset in the experiment with significant results.

In this paper, we investigate the distribution of classification accuracies of multiclass problems in the feature space and analyze performances of the conventional feature extraction algorithms. In order to find the distribution of classification accuracies, we sample the feature space and compute the classification accuracy corresponding to each sampling point. Experimental results showed that there exist much better feature sets that the conventional feature extraction algorithms fail to find. In addition, the distribution of classification accuracies is useful for developing and evaluating the feature extraction algorithm.

최근 활발히 연구가 진행 중인 유전발현 데이타를 이용한 다중클래스 암 분류는 DNA 마이크로어레이로부터 획득된 대규모의 유전자 정보를 분석하여 암의 종류를 판단한다. 수집된 유전발현 데이타에는 대상 암과 관련이 없는 유전자도 포함되어 있기 때문에 높은 성능의 분류 결과를 얻기 위해서 유용한 유전자를 선택하는 것이 필요하다. 기존의 순위기반 유전자 선택은 이진클래스를 대상으로 고안되었고 이상표식 유전자(Ideal marker gene)를 이용하기 때문에 다중클래스 암 분류에 직접 적용하기에는 한계가 있다. 본 논문에서는 이상표식 유전자를 사용하지 않고 유전발현 수준의 분포를 직접 분석하는 순위기반 다중클래스 유전자 선택 기법을 제안한다. 유전발현 수준을 이산화하고 학습 데이타로부터 빈도를 계산하여 클래스 간 분별력을 측정한 후, 선택된 유전자를 이용하여 나이브 베이즈 분류기를 사용해 다중 암 분류를 수행한다. 제안하는 방법을 다수의 다중클래스 암 분류 데이타에 적용하여 기존 유전자 선택 방법에 비해 우수함을 확인하였다.

Multiclass cancer classification has been actively investigated based on gene expression profiles, where it determines the type of cancer by analyzing the large amount of gene expression data collected by the DNA microarray technology. Since gene expression data include many genes not related to a target cancer, it is required to select informative genes in order to obtain highly accurate classification. Conventional rank-based gene selection methods often use ideal marker genes basically devised for binary classification, so it is difficult to directly apply them to multiclass classification. In this paper, we propose a novel method for multiclass gene selection, which does not use ideal marker genes but directly analyzes the distribution of gene expression. It measures the class-discriminability by discretizing gene expression levels into several regions and analyzing the frequency of training samples for each region, and then classifies samples by using the naive Bayes classifier. We have demonstrated the usefulness of the proposed method for various representative benchmark datasets of multiclass cancer classification.