Academic literature on the topic 'Nu-svm'

It is crucial to study the axial compression behavior of concrete-filled steel tubular (CFST) columns to ensure the safe operation of engineering structures. The restriction between steel tubular and core concrete in CFSTs is complex and the relationship between geometric and material properties and axial compression behavior is highly nonlinear. These challenges have prompted the use of soft computing methods to predict the ultimate bearing capacity (abbreviated as Nu) under axial compression. Taking the square CFST short column as an example, a mass of experimental data is obtained through axial compression tests. Combined with support vector machine (SVM) and particle swarm optimization (PSO), this paper presents a new method termed PSVM (SVM optimized by PSO) for Nu value prediction. The nonlinear relationship in Nu value prediction is efficiently represented by SVM, and PSO is used to select the model parameters of SVM. The experimental dataset is utilized to verify the reliability of the PSVM model, and the prediction performance of PSVM is compared with that of traditional design methods and other benchmark models. The proposed PSVM model provides a better prediction of the ultimate axial capacity of square CFST short columns. As such, PSVM is an efficient alternative method other than empirical and theoretical formulas.

ABSTRACT The study of machine learning (ML) techniques for the autonomous classification of astrophysical sources is of great interest, and we explore its applications in the context of a multifrequency data-frame. We test the use of supervised ML to classify blazars according to its synchrotron peak frequency, either lower or higher than 1015 Hz. We select a sample with 4178 blazars labelled as 1279 high synchrotron peak (HSP: $\rm \nu$-peak > 1015 Hz) and 2899 low synchrotron peak (LSP: $\rm \nu$-peak < 1015 Hz). A set of multifrequency features were defined to represent each source that includes spectral slopes ($\alpha _{\nu _1, \nu _2}$) between the radio, infra-red, optical, and X-ray bands, also considering IR colours. We describe the optimization of five ML classification algorithms that classify blazars into LSP or HSP: Random forests (RFs), support vector machine (SVM), K-nearest neighbours (KNN), Gaussian Naive Bayes (GNB), and the Ludwig auto-ML framework. In our particular case, the SVM algorithm had the best performance, reaching 93 per cent of balanced accuracy. A joint-feature permutation test revealed that the spectral slopes alpha-radio-infrared (IR) and alpha-radio-optical are the most relevant for the ML modelling, followed by the IR colours. This work shows that ML algorithms can distinguish multifrequency spectral characteristics and handle the classification of blazars into LSPs and HSPs. It is a hint for the potential use of ML for the autonomous determination of broadband spectral parameters (as the synchrotron ν-peak), or even to search for new blazars in all-sky data bases.

Objective. Immune checkpoint inhibitors, such as programmed death-1/ligand-1 (PD-1/L1), exhibited autoimmune-like disorders, and hyperglycemia was on the top of grade 3 or higher immune-related adverse events. Machine learning is a model from past data for future data prediction. From post-marketing monitoring, we aimed to construct a machine learning algorithm to efficiently and rapidly predict hyperglycemic adverse reaction in patients using PD-1/L1 inhibitors. Methods. In original data downloaded from Food and Drug Administration Adverse Event Reporting System (US FAERS), a multivariate pattern classification of support vector machine (SVM) was used to construct a classifier to separate adverse hyperglycemic reaction patients. With correct core SVM function, a 10-fold 3-time cross validation optimized parameter value composition in model setup with R language software. Results. The SVM prediction model was set up from the number type/number optimization method, as well as the kernel and type of “rbf” and “nu-regression” composition. Two key values (nu and gamma) and case number displayed high adjusted r2 in curve regressions ( n u = 0.5649 × e − case / 6984 , gamma = 9.005 × 10 − 4 × case − 4.877 × 10 − 8 × case 2 ). This SVM model with computable parameters greatly improved the assessing indexes (accuracy, F1 score, and kappa) as well as coequal sensitivity and the area under the curve (AUC). Conclusion. We constructed an effective machine learning model based on compositions of exact kernels and computable parameters; the SVM prediction model can noninvasively and precisely predict hyperglycemic adverse drug reaction (ADR) in patients treated with PD-1/L1 inhibitors, which could greatly help clinical practitioners to identify high-risk patients and perform preventive measurements in time. Besides, this model setup process provided an analytic conception for promotion to other ADR prediction, such ADR information is vital for outcome improvement by identifying high-risk patients, and this machine learning algorithm can eventually add value to clinical decision making.

У статті розглянуто динаміку та поведінку кіберспортивної індустрії на світовому рівні та стан кіберспорту як індустрії в Україні. Визначено основні досягнення кіберспортивної сфери України. Проведено статистичний аналіз доходу кіберспортивної індустрії, сукупної аудиторії кіберспортивних ігор, постійних та пересічних глядачів змагань на основі аналізу варіації, моди, показників асиметрії та ексцесу розподілу. Для досягнення цілей дослідження було використано метод експоненційного згладження та метод опорних векторів. SVM є методом машинного навчання, який використовується для вирішення задач класифікації та регресії. Як і для класичної моделі регресії основою підходу є знаходження функції підгонки емпіричних даних. Обрані методи дозволили підготувати дані для аналізу та побудувати регресійні SVM-моделі з ядром на основі радіально-базисних функцій. Побудовані моделі для доходу кіберспорту та пересічних глядачів кіберспорту мають тип epsilon-SVM, а для світової аудиторії кіберспорту та постійних глядачів кіберспорту – nu-SVM. Доведена адекватність побудованих моделей на основі аналізу залишків моделі. Здійснено прогнозування вхідних показників. Визначено, що до 2025 року очікується постійне зростання доходу від кіберспортивної діяльності, що означає постійний розвиток та вдосконалення супутньої до кіберспорту інфраструктури. Визначено важливість та необхідність державної підтримки розвитку кіберспорту на всіх рівнях: від організації турнірних площадок до проведення регіональних, шкільних, аматорських турнірів. Отримані результати можуть бути використані Федерацією кіберспорту України, кіберспортивними організаціями, дослідниками для обгрунтування необхідності розвитку кіберспорту в Україні.

Support Vector Machines (SVM) are widely used in many fields of science, including system identification. The selection of feature vector plays a crucial role in SVM-based model building process. In this paper, we investigate the influence of the selection of feature vector on model’s quality. We have built an SVM model with a non-linear ARX (NARX) structure. The modelled system had a SISO structure, i.e. one input signal and one output signal. The output signal was temperature, which was controlled by a Peltier module. The supply voltage of the Peltier module was the input signal. The system had a non-linear characteristic. We have evaluated the model’s quality by the fit index. The classical feature selection of SVM with NARX structure comes down to a choice of the length of the regressor vector. For SISO models, this vector is determined by two parameters: nu and ny. These parameters determine the number of past samples of input and output signals of the system used to form the vector of regressors. In the present research we have tested two methods of building the vector of regressors, one classic and one using custom regressors. The results show that the vector of regressors obtained by the classical method can be shortened while maintaining the acceptable quality of the model. By using custom regressors, the feature vector of SVM can be reduced, which means also the reduction in calculation time.

In this study, we measured the convergence rate using the mean-squared error (MSE) of the standardized neuropsychological test to determine the severity of Parkinson’s disease dementia (PDD), which is based on support vector machine (SVM) regression (SVR) and present baseline data in order to develop a model to predict the severity of PDD. We analyzed 328 individuals with PDD who were 60 years or older. To identify the SVR with the best prediction power, we compared the classification performance (convergence rate) of eight SVR models (Eps-SVR and Nu-SVR with four kernel functions (a radial basis function (RBF), linear algorithm, polynomial algorithm, and sigmoid)). Among the eight models, the MSE of Nu-SVR-RBF was the lowest (0.078), with the highest convergence rate, whereas the MSE of Eps-SVR-sigmoid was 0.110, with the lowest convergence rate. The results of this study imply that this approach could be useful for measuring the severity of dementia by comprehensively examining axial atypical features, the Korean instrumental activities of daily living (K-IADL), changes in rapid eye movement sleep behavior disorder (RBD), etc. for optimal intervention and caring of the elderly living alone or patients with PDD residing in medically vulnerable areas.

Real-time bus travel time prediction has been an interesting problem since past decade, especially in India. Popular methods for travel time prediction include time series analysis, regression methods, Kalman filter method and Artificial Neural Network (ANN) method. Reported studies using these methods did not consider the high variance situations arising from the varying traffic and weather conditions, which is very common under heterogeneous and lane-less traffic conditions such as the one in India. The aim of the present study is to analyse the variance in bus travel time and predict the travel time accurately under such conditions. Literature shows that Support Vector Machines (SVM) technique is capable of performing well under such conditions and hence is used in this study. In the present study, nu-Support Vector Regression (SVR) using linear kernel function was selected. Two models were developed, namely spatial SVM and temporal SVM, to predict bus travel time. It was observed that in high mean and variance sections, temporal models are performing better than spatial. An algorithm to dynamically choose between the spatial and temporal SVM models, based on the current travel time, was also developed. The unique features of the present study are the traffic system under consideration having high variability and the variables used as input for prediction being obtained from Global Positioning System (GPS) units alone. The adopted scheme was implemented using data collected from GPS fitted public transport buses in Chennai (India). The performance of the proposed method was compared with available methods that were reported under similar traffic conditions and the results showed a clear improvement.

Image processing techniques are widely used for the detection and classification of diseases for various plants. The structure of the plant and appearance of the disease on the plant pose a challenge for image processing. This research implements SVM (Support Vector Machine) based image-processing approach to analyze and classify three of the rice crop diseases. The process consists of two phases, i.e. training phase and disease prediction phase. The approach identifies disease on the leaf using trained classifier. The proposed research work optimizes SVM parameters (gamma, nu) for maximum efficiency. The results show that the proposed approach achieved 94.16% accuracy with 5.83% misclassification rate, 91.6% recall rate and 90.9% precision. These findings were compared with image processing techniques discussed in review of literature. The results of comparison conclude that the proposed methodology yields high accuracy percentage as compared to the other techniques. The results obtained can help the development of an effective software solution by incorporating image processing and collaboration features. This may facilitate the farmers and other bodies in effective decision making to efficiently protect the rice crops from substantial damage. While considering the findings of this research, the presented technique may be considered as a potential solution for adding image processing techniques to KM (Knowledge Management) systems.

Dans cet article, nous évaluons les performances de classification de trois algorithmes non paramétriques (kNN, RF et SVM) en utilisant les données multi-temporelles de trois satellites (Sentinel-1, Alos-Palsar-2 et Sentinel-2) et de leurs combinaisons. La zone d'étude choisie se caractérise par un climat méditerranéen subhumide et une topographie très accidentée qui rend la classification d’occupation du sol particulièrement difficile. En outre, elle contient une aire protégée nommée Jbel Moussa et présente une diversité biologique exceptionnelle. Afin de suivre le couvert végétal de cette dernière, nous avons acquis et prétraités les images satellitaires optiques et radar pour la période du 1er janvier au 31 décembre 2017. Ensuite, nous avons combiné les trois satellites, soit douze scénarios produits. Des cartes de classifications illustrent notre approche. Un total de trente-six classifications a été obtenu, en se basant sur sept classes : eau, bâtiment et infrastructures, sol nu, végétation peu dense, prairies, forêt peu dense et forêt dense. Les résultats ont montré que pour tous les scénarios, la précision globale la plus élevée a été produite par RF (53,03%-93,06%), suivie de kNN (49,16%-89,63%), tandis que SVM (47,86%-86,08%) a produit la précision de classification la plus faible. L'étude a également montré une similitude entre les performances de la combinaison des trois satellites et celles de Sentinel-2 seul. Les estimations de la superficie pour les différentes classes vont de 0,85 km2 (0,11% de la zone d'étude) à 326,84 km2 (41,31% de la zone d'étude)

The Support Vector Machine (SVM) is a binary classification paradigm based on statistical learning. It is an important tool in object detection and pattern recognition, as well as signal processing, and the SVM algorithm has been shown to outperform existing classification algorithms. There are two main drawbacks with the original SVM formulation: the classification performance of the tuning regularisation parameter, and the computational and time cost of training phase. This Thesis focuses on the development of a new formulation of 2v-SVM, with two cost parameters v₊ and v₋, to mitigate these drawbacks. In addition, the computation implementation of training process is designed and explained, and the relationship between the new and the original formulations is developed mathematically and discussed. As with many statistical learning problems, the formulation of the original SVM (C-SVM) uses a regularisation parameter (C) to balance the generalisation performance of the classifier. The regularisation parameter places equal weightings on the number of training errors of both class labels. The equal weightings can be undesirable, and can be detrimental to the overall performance of the resulting classifier. In this Thesis, SVMs are extended to dual-error parameters (2C-SVMs and 2v-SVMs) that improve the classification performance, in particularly where the training class sizes are greatly different, and allow classification biasing based on a priori information. We discuss the creation of the new formulation and show results that provide indications of the improvements, and illustrate the use of 2v-SVMs for multi-category classification. In addition, we describe a metric that measures the reliability of a multi-category classification for the one-against-rest strategy. We introduce a novel implementation for training 2v-SVMs. The weakness of SVMs is the computation cost of training multiple SVMs to produce an optimised classifier. Each training involves the optimisation of a quadratic programming problem that is non-trivial. Optimisation algorithms for C-SVMs have proved to reduce the computational cost involved, but these algorithms cannot be directly applied to 2v-SVMs. The novel implementation improves on the existing methods while reducing the computational requirements. The relationship between the new 2v-SVM formulation and the original describes the link between the objective functions of the two formulations. The mathematical link shows that the new formulation provides the same functionality of the original formulation, while improving on the classification and computation performances. The resulting extension and implementation of 2v-SVMs provides a strong and robust method for producing binary classifiers as well as multi-category classifiers. The research detailed here has resulted in 2v-SVMs that have improved the computation and classification performance of SVMs. These improvements contribute to the broad area of multi-dimensional signal processing, such as for signal detection and image classification. The robustness of the 2v-SVM in forming working classifiers from unprocessed data, and the reduction of number of training cycles allows users to quickly and effectively produce results for their classification problems.
Thesis (Ph.D.) -- University of Adelaide, School of Electrical and Electronic Engineering, 2013