Multi-level Support Vector Machine Research Articles

Development of a Generalized Multilevel SVM and Capacitor Voltage Balancing Strategy for Multiphase Three-Level NPC Converters

Development of a Generalized Multilevel SVM and Capacitor Voltage Balancing Strategy for Multiphase Three-Level NPC Converters

A multi-level decision-making framework for heart-related disease prediction and recommendation

The precise prediction of health-related issues is a significant challenge in healthcare, with heart-related diseases posing a particularly threatening global health problem. Accurate prediction and recommendation for heart-related diseases are crucial for timely and effective treatment solutions. The primary objective of this study is to develop a classification model capable of accurately identifying heart diseases and providing appropriate recommendations for patients. The proposed system utilizes a multilevel-based classification mechanism employing Support Vector Machines. It aims to categorize heart diseases by analyzing patient’s vital parameters. The performance of the proposed model was evaluated by testing it on a dataset containing patient records. The generated recommendations are based on a comprehensive assessment of the severity of clinical features exhibited by patients, including estimating the associated risk of both clinical features and the disease itself. The predictions were evaluated using three metrics: accuracy, specificity, and the receiver operating characteristic curve. The proposed Multilevel Support Vector Machine (MSVM) classification model achieved an accuracy rate of 94.09% in detecting the severity of heart disease. This makes it a valuable tool in the medical field for providing timely diagnosis and treatment recommendations. The proposed model presents a promising approach for accurately predicting heart-related diseases and highlights the potential of soft computing techniques in healthcare. Future research could focus on further enhancing the proposed model’s accuracy and applicability.

Introducing an Automatic Bread Quality Assessment Algorithm using Image Processing Techniques

In this research, an automatic algorithm of bread quality assessment using image processing techniques, is proposed. First, color images of bread with different qualities are photographed and a database of 1250 bread images is prepared. Then 2320 color and texture features are extracted from each bread images. Then, from this number of features, only 15 features containing sufficient information are selected. In addition, 54 appearance features are extracted from each bread image to determine its shape and size. Finally, bread images are classified using the multilevel Support Vector Machine classifier. The classification process is divided into five "one-against-all" classification problems. The proposed algorithm correctly identifies the bread appearance defects, including cuts, fractures, folds, non-uniformity, black and burnt areas in baking, deformity, color and size. The proposed algorithm, considering the extraction of only 15 features per an image, has a speed that guarantees its use in a machine vision system. The performance success of the proposed algorithm on the bread database, despite its very simple implementation, is 96.95%.

A novel graph search and machine learning method to detect and locate high impedance fault zone in distribution system

AbstractHigh impedance fault (HIF) is difficult to detect by conventional overcurrent protection relays due to the lower fault current values, which are normally lower than the normal current. A fast and reliable algorithm is required to detect this type of fault. This paper proposes a novel method for detecting the location of HIF fault zone in a distribution system by using a novel graph theory‐based zone detection technique along with a Random Search Multilevel Support Vector Machine (RSMSVM) algorithm to classify the faulted zone. Due to shift in‐variance property of “Dual Tree Complex Wavelet Transform (DTCWT),” which has been used, in this paper, to decompose the voltage/current waveform to collect the signature of the signals and feed to the optimized RSMSVM model for classifying fault zone. The proposed method is evaluated on the IEEE 33‐bus system and also IEEE 39 bus test system under normal and noisy conditions. The proposed method is also evaluated for distribution network with the integration of distributed generation.

INVENTORY OF HIGH VALUE CROPS USING LIDAR DATA AND GIS IN LANAO DEL NORTE PHILIPPINES

One of the objectives of Republic Act 8435 or Philippines Agriculture and Fisheries Modernization Act of 1997 is to modernize the agriculture and fisheries sectors by developing it into a technology –supported division. In the Philippines, about 32% of the country's total land areas were agricultural lands. Mapping and delineating the agricultural crops are significant in the implementation of this Act. Moreover, having an updated and accurate inventory of the agricultural resources in an area is critical in the assessment, planning, and development specifically in providing the necessary government support to improve the agricultural system, like Farm-to-Market Roads (FMR), irrigation system, pre- and post-harvest facilities, among others. This study extracted agricultural crops like rice, corn, coconut, banana, and mango present in the province of Lanao Del Norte using Light Detection and Ranging (LiDAR) technology and Geographic Information System (GIS). LiDAR is an active remote sensing system with an illumination source from laser lights. Object-based image analysis (OBIA), multi-level support vector machine (SVM) classifier, and ground truthing were the general procedure in extracting and validating these crops. OBIA allows the segmentation of images derived from LiDAR datasets into meaningful objects. SVM classifier analyzes and identifies the pattern in the attributes of the segmented objects and classifies these objects based on its analysis. 71% of the extracted agricultural areas are planted with coconut, followed by banana with 13%, corn with 8%, rice constitutes 5%, and lastly mango with 3% of the total agricultural area. These extracted crops can now be utilized in updating the agricultural resource accounting of the province. Accuracy assessment was performed using 3,300 validation points gathered from the field and from orthorectified photos. Result of evaluating the accuracy shows that 80-90% of the crops were extracted and classified correctly. The accuracy assessment records an overall accuracy of 0.915 and a Kappa Index of Agreement (KIA) of 0.89147

Music emotion recognition method based on multi feature fusion

There are some problems in music emotion recognition, such as large root mean square error of recognition results and low Pearson correlation coefficient. The music signal is divided into frames by window function, the noise in the music signal is reduced by the time domain endpoint detection, and the music signal is preprocessed. The characteristics of pitch change, gene rise and fall, speech speed and gene slope were extracted by Mehr frequency cepstrum coefficient. According to the extracted music emotion features, the multi-feature fusion kernel function is constructed. Based on the fusion results, the multi-level SVM emotion recognition model is built with the support vector mechanism to realise music emotion recognition. Experimental results show that the root mean square error of the proposed method is always within the range of 0.02, and the highest Pearson correlation coefficient is about 0.9.

Multilevel SVM and AI based Transformer Fault Diagnosis using the DGA Data

The Dissolved Gas Analysis (DGA) is utilized as a test for the detection of incipient problems in transformers, and condition monitoring of transformers using software-based diagnosis tools has become crucial. This research uses dissolved gas analysis as an intelligent fault classification of a transformer. The Multilayer SVM technique is used to determine the classification of faults and the name of the gas. The learned classifier in the multilayer SVM is trained with the training samples and can classify the state as normal or fault state, which contains six fault categories. In this paper, polynomial and Gaussian functions are utilized to assess the effectiveness of SVM diagnosis. The results demonstrate that the combination ratios and graphical representation technique is more suitable as a gas signature, and that the SVM with the Gaussian function outperforms the other kernel functions in diagnosis accuracy.

Faster Support Vector Machines

The time complexity of support vector machines (SVMs) prohibits training on huge datasets with millions of data points. Recently, multilevel approaches to train SVMs have been developed to allow for time-efficient training on huge datasets. While regular SVMs perform the entire training in one—time-consuming—optimization step, multilevel SVMs first build a hierarchy of problems decreasing in size that resemble the original problem and then train an SVM model for each hierarchy level, benefiting from the solved models of previous levels. We present a faster multilevel support vector machine that uses a label propagation algorithm to construct the problem hierarchy. Extensive experiments indicate that our approach is up to orders of magnitude faster than the previous fastest algorithm while having comparable classification quality. For example, already one of our sequential solvers is on average a factor 15 faster than the parallel ThunderSVM algorithm, while having similar classification quality. 1

AAR-LN-DQ: Automatic anatomy recognition based disease quantification in thoracic lymph node zones via FDG PET/CT images without Nodal Delineation.

The derivation of quantitative information from medical images in a practical manner is essential for quantitative radiology (QR) to become a clinical reality, but still faces a major hurdle because of image segmentation challenges. With the goal of performing disease quantification in lymph node (LN) stations without explicit nodal delineation, this paper presents a novel approach for disease quantification (DQ) by automatic recognition of LN zones and detection of malignant lymph nodes within thoracic LN zones via positron emission tomography/computed tomography (PET/CT) images. Named AAR-LN-DQ, this approach decouples DQ methods from explicit nodal segmentation via an LN recognition strategy involving a novel globular filter and a deep neural network called SegNet. The methodology consists of four main steps: (a) Building lymph node zone models by automatic anatomy recognition (AAR) method. It incorporates novel aspects of model building that relate to finding an optimal hierarchy for organs and lymph node zones in the thorax. (b) Recognizing lymph node zones by the built lymph node models. (c) Detecting pathologic LNs in the recognized zones by using a novel globular filter (g-filter) and a multi-level support vector machine (SVM) classifier. Here, we make use of the general globular shape of LNs to first localize them and then use a multi-level SVM classifier to identify pathologic LNs from among the LNs localized by the g-filter. Alternatively, we designed a deep neural network called SegNet which is trained to directly recognize pathologic nodes within AAR localized LN zones. (d) Disease quantification based on identified pathologic LNs within localized zones. A fuzzy disease map is devised to express the degree of disease burden at each voxel within the identified LNs to simultaneously handle several uncertain phenomena such as PET partial volume effects, uncertainty in localization of LNs, and gradation of disease content at the voxel level. We focused on the task of disease quantification in patients with lymphoma based on PET/CT acquisitions and devised a method of evaluation. Model building was carried out using 42 near-normal patient datasets via contrast-enhanced CT examinations of their thorax. PET/CT datasets from an additional 63 lymphoma patients were utilized for evaluating the AAR-LN-DQ methodology. We assess the accuracy of the three main processes involved in AAR-LN-DQ via fivefold cross validation: lymph node zone recognition, abnormal lymph node localization, and disease quantification. The recognition and scale error for LN zones were 12.28mm±1.99 and 0.94±0.02, respectively, on normal CT datasets. On abnormal PET/CT datasets, the sensitivity and specificity of pathologic LN recognition were 84.1%±0.115 and 98.5%±0.003, respectively, for the g-filter-SVM strategy, and 91.3%±0.110 and 96.1%±0.016, respectively, for the SegNet method. Finally, the mean absolute percent errors for disease quantification of the recognized abnormal LNs were 8%±0.09 and 14%±0.10 for the g-filter-SVM method and the best SegNet strategy, respectively. Accurate disease quantification on PET/CT images without performing explicit delineation of lymph nodes is feasible following lymph node zone and pathologic LN localization. It is very useful to perform LN zone recognition by AAR as this step can cover most (95.8%) of the abnormal LNs and drastically reduce the regions to search for abnormal LNs. This also improves the specificity of deep networks such as SegNet significantly. It is possible to utilize general shape information about LNs such as their globular nature via g-filter and to arrive at high recognition rates for abnormal LNs in conjunction with a traditional classifier such as SVM. Finally, the disease map concept is effective for estimating disease burden, irrespective of how the LNs are identified, to handle various uncertainties without having to address them explicitly one by one.

Rice disease classification based on leaf image using multilevel Support Vector Machine (SVM)

Plant disease is one of many factors that decrease the quality and quantity value of agriculture, especially rice plants. Automatic technology based on digital image processing is being developed to overcome this problem. Support Vector Machine (SVM) is one of the most used classifications and detection methods. SVM has been developed into multi SVM by combining several binary SVMs to classify more than two classes. In the proposed system, we use one of the multi SVM strategy, namely One Vs. All. The accuracy of classification reaches 86.10% using linear kernel. It has a higher value of accuracy than using polynomial and RBF kernel function. The scenario for the number of the dataset used is 70% for the training set and 30% for the testing set from a whole 240 images.

Detection and Classification of Defects Using ECT and Multi-Level SVM Model

In this work, classification of cracks in a non-ferromagnetic material (aluminum 1050) is presented using two supervised machine learning techniques called support vector machines (SVM) and decision tree. The classification performances between the two classifiers were evaluated using confusion matrices and miss-classification errors. In order to improve the classification performance, a multi-level SVM model is proposed wherein each level implements SVM using different features and kernels. The experimental tests were performed in two aluminum plates containing 18 defects with different depths and lengths. The 18 defects were inspected in the surface and sub-surface locations, resulting in a study of 36 different crack cases. The experimental data used to extract features for supervised machine learning were obtained using an eddy current testing (ECT) probe that includes a planar coil and a tunnel magnetoresistance (TMR) sensor. The probe was excited by a multi-frequency signal with four testing frequencies to evaluate the dimensions and the locations of the defects. Due to the experimental noise existing in the measurement signals, it is important to pre-process the signals before feature extraction. Signal processing techniques (such as moving average filter, spline interpolation and symmetry corrections) were used before extracting the features for classification.

Extended Local Binary Pattern Features based Face Recognition using Multilevel SVM Classifier

The Face recognition method is one of the authoritative biometric system in recognition methods to recognize the individual, because face is a distinctive biometric trait of an human being and it is the superior method of recognition. This paper proposes a novel Face recognition method by using extended LBP features. The pre-processing is carried out to extract the face area using viola-johns algorithm and all images are resized to 100x100. The LBP operator is applied on resized face images by rotating the each image by 15 degrees, i.e., at 7 degree left and 7 degree right and at zero degree to extract the feature vectors and final features are obtained by applying histogram technique. The SVM classifier is used for matching the database images with test images to measure the performance such as TSR, FAR, FRR & EER. The performance parameters are compared with existing algorithms for YALE and FERET database.

Higher Order Feature Extraction and Selection for Robust Human Gesture Recognition using CSI of COTS Wi-Fi Devices.

Device-free human gesture recognition (HGR) using commercial off the shelf (COTS) Wi-Fi devices has gained attention with recent advances in wireless technology. HGR recognizes the human activity performed, by capturing the reflections of Wi-Fi signals from moving humans and storing them as raw channel state information (CSI) traces. Existing work on HGR applies noise reduction and transformation to pre-process the raw CSI traces. However, these methods fail to capture the non-Gaussian information in the raw CSI data due to its limitation to deal with linear signal representation alone. The proposed higher order statistics-based recognition (HOS-Re) model extracts higher order statistical (HOS) features from raw CSI traces and selects a robust feature subset for the recognition task. HOS-Re addresses the limitations in the existing methods, by extracting third order cumulant features that maximizes the recognition accuracy. Subsequently, feature selection methods derived from information theory construct a robust and highly informative feature subset, fed as input to the multilevel support vector machine (SVM) classifier in order to measure the performance. The proposed methodology is validated using a public database SignFi, consisting of 276 gestures with 8280 gesture instances, out of which 5520 are from the laboratory and 2760 from the home environment using a 10 × 5 cross-validation. HOS-Re achieved an average recognition accuracy of 97.84%, 98.26% and 96.34% for the lab, home and lab + home environment respectively. The average recognition accuracy for 150 sign gestures with 7500 instances, collected from five different users was 96.23% in the laboratory environment.

Engineering fast multilevel support vector machines

The computational complexity of solving nonlinear support vector machine (SVM) is prohibitive on large-scale data. In particular, this issue becomes very sensitive when the data represents additional difficulties such as highly imbalanced class sizes. Typically, nonlinear kernels produce significantly higher classification quality to linear kernels but introduce extra kernel and model parameters which requires computationally expensive fitting. This increases the quality but also reduces the performance dramatically. We introduce a generalized fast multilevel framework for regular and weighted SVM and discuss several versions of its algorithmic components that lead to a good trade-off between quality and time. Our framework is implemented using PETSc which allows an easy integration with scientific computing tasks. The experimental results demonstrate significant speed up compared to the state-of-the-art nonlinear SVM libraries. Reproducibility: our source code, documentation and parameters are available at https://github.com/esadr/mlsvm .

Multi Intelligent Fuzzy Integration (Mifi) Support Vector Machines for the Mammogram Classification

Breast cancer is the most important problem across the globe in which the 80% of the women are suffering without knowing the causes and effects of the cancer cells. Mammogram Image is the most powerful tool for the diagnosis of the Breast cancer. The analysis of this mammogram images proves to be more vital in terms of diagnosis but the accuracy level still needs improvisation. Several intelligent techniques are suggested for the detection of Micro calcification in mammogram images. The new technique MIFI-SVM has been proposed which integrates the GLCM features along with the Fuzzy Support Vector Machines. ROI Segmentation using Saliency maps has been used for the proposed algorithm and feature is extracted using GLCM and fed to Fuzzy Support Vector Machines The MIAS datasets has been used for testing the proposed algorithm and accuracy, sensitivity has been measured which proves to be better when compared to other Multi-level SVM’s, C-SVM and Neural Networks.

Real-time multi-agent system for an adaptive intrusion detection system

An adaptive intrusion detection system that can detect unknown attacks in real-time network traffic is a major concern. Conventional adaptive intrusion detection systems are computationally expensive in terms of computer resources and time because these systems have to be retrained with known and unknown attacks. In this study, a method called Real-Time Multi-agent System for an Adaptive Intrusion Detection System RTMAS-AIDS, which is based on a multi-agent system, is proposed to allow the intrusion detection system to adapt to unknown attacks in real-time. This method utilizes the classification models multi-level hybrid SVM and ELM to detect normal behavior and known attacks. An adaptive SVM model, in which processes run in parallel and are distributed in MAS, is also used to detect and learn new attacks in real-time. Results show that the proposed method significantly reduced the training cost of detecting unknown attacks compared with the conventional method. In addition, the analysis results of the popular KDDCup’99 dataset show that RTMAS-AIDS can detect Probe, R2L, and U2R attacks better than the non-retrained multi-agent system using the multi-level hybrid SVM and ELM models as well as the multi-level hybrid SVM and ELM. RTMAS-AIDS exhibited a significantly improved detection accuracy that reached 95.86% and can detect and learn unknown attacks faster (up to 61%) than the other two methods (MAS-MLSE and MLSE).

Perceptron ensemble of graph-based positive-unlabeled learning for disease gene identification

Identification of disease genes, using computational methods, is an important issue in biomedical and bioinformatics research. According to observations that diseases with the same or similar phenotype have the same biological characteristics, researchers have tried to identify genes by using machine learning tools. In recent attempts, some semi-supervised learning methods, called positive-unlabeled learning, is used for disease gene identification. In this paper, we present a Perceptron ensemble of graph-based positive-unlabeled learning (PEGPUL) on three types of biological attributes: gene ontologies, protein domains and protein-protein interaction networks. In our method, a reliable set of positive and negative genes are extracted using co-training schema. Then, the similarity graph of genes is built using metric learning by concentrating on multi-rank-walk method to perform inference from labeled genes. At last, a Perceptron ensemble is learned from three weighted classifiers: multilevel support vector machine, k-nearest neighbor and decision tree. The main contributions of this paper are: (i) incorporating the statistical properties of gene data through choosing proper metrics, (ii) statistical evaluation of biological features, and (iii) noise robustness characteristic of PEGPUL via using multilevel schema. In order to assess PEGPUL, we have applied it on 12950 disease genes with 949 positive genes from six class of diseases and 12001 unlabeled genes. Compared with some popular disease gene identification methods, the experimental results show that PEGPUL has reasonable performance.

Detection and localization of damage using empirical mode decomposition and multilevel support vector machine

Damage in the structure may raise a significant amount of maintenance cost and serious safety problems. Hence detection of the damage at its early stage is of prime importance. The main contribution pursued in this investigation is to propose a generic optimal methodology to improve the accuracy of positioning of the flaw in a structure. This novel approach involves a two-step process. The first step essentially aims at extracting the damage-sensitive features from the received signal, and these extracted features are often termed the damage index or damage indices, serving as an indicator to know whether the damage is present or not. In particular, a multilevel SVM (support vector machine) plays a vital role in the distinction of faulty and healthy structures. Formerly, when a structure is unveiled as a damaged structure, in the subsequent step, the position of the damage is identified using Hilbert–Huang transform. The proposed algorithm has been evaluated in both simulation and experimental tests on a 6061 aluminum plate with dimensions 300 mm × 300 mm × 5 mm which accordingly yield considerable improvement in the accuracy of estimating the position of the flaw.

A Novel Space Vector Technique for the Direct Three-level Matrix Converter

This study proposes a novel Direct Torque Control (DTC) method for the Direct Three level Matrix Converter (DTMC), which uses both the input phase voltage vectors (short vectors) and the input line voltage vectors (long vectors). The problem of voltage imbalance at the input filter capacitance due to the use of the short vectors is addressed with an additional voltage hysteresis comparator. With the errors of torque, flux, sin Ψ and the neutral point voltage, an Optimum Switching Table (OST) is designed for the DTMC. The OST generates the necessary switching signals for the DTC of the DTMC. The DTMC topology with the modified ISVM technique reduces the THD at the output. The proposed DTMC Indirect Space Vector pulse width Modulation (ISVM) technique uses the idea of multilevel inverter SVM technique along with the proposed neutral current balancing strategy for generating the firing pulses. The switching loss model for the DTMC is developed and the performance of the DTMC is compared with that of the Conventional Matrix Converter (CMC). The performance of the proposed DTC technique for the DTMC is evaluated through simulation to explain the reduced torque ripple characteristics. To validate the proposed DTMC ISVM technique, a 3 kVA direct multilevel matrix converter prototype was developed.

질감 분석을 이용한 유도 전동기의 기계적 결함 분류

본 논문에서는 유도 전동기의 기계적 결함을 진단하기 위해 진동신호와 질감 분석을 이용한 알고리즘을 제안한다. 영상화된 결함 신호가 갖는 무늬, 색상 대비의 특징을 분석하고, 그레이레벨 동시발생행렬(Gray-Level Co-occurrence Model, GLCM)을통해 세 가지 질감특징을추출한다. 추출된 세 가지질감 특징을 RBF(Radial Basis Function) 커널 함수를 사용하는 다중레벨 서포터 벡터 머신(Multi-Level Support Vector Machine, MLSVM)의 입력으로 사용하여 결함 유형을 분류한다. 결함 유형을 분류하는 최적의 MLSVM을 위한 RBF 커널 함수의 매개변수를 찾기 위해 매개변수 값을 0.3부터 1.0으로 바꿔가며 분류성능을 평가한 결과, 결함 유형별로 0.3에서 0.6사이의 매개변수 값에서 100%에 가까운 분류 정확성을 보였다. 또한 15dB, 20dB의 잡음이 첨가된 진동신호를 이용한 실험에서도 평균 98%이상의 높은 분류 정확성을 보였다. This paper proposes an algorithm using vibration signals and texture analysis for mechanical fault diagnosis of an induction motor. We analyze characteristics of contrast and pattern of an image converted from vibration signal and extract three texture features using gray-level co-occurrence model(GLCM). Then, the extracted features are used as inputs of a multi-level support vector machine(MLSVM) which utilizes the radial basis function(RBF) kernel function to classify each fault type. In addition, we evaluate the classification performance with varying the parameter from 0.3 to 1.0 for the RBF kernel function of MLSVM, and the proposed algorithm achieved 100% classification accuracy with the parameter of the RBF from 0.3 to 1.0. Moreover, the proposed algorithm achieved about 98% classification accuracy with 15dB and 20dB noise inserted vibration signals.