Quadratic Support Vector Machine Research Articles

An efficient machine learning approach for activity recognition

The population of older people in western nations are growing drastically. Independent lifestyle is their preference and this leads to fall instances often. Falls of this kind results in severe health issues or sometimes causes deadly damages to the elderly people. Considering this problem, it is highly important to come up with fall discovery systems. A machine learning framework has been proposed in this regard which covers both fall discovery and day to day movement recognition. Utilizing acceleration and speed inputs from dual unrestricted repositories are highly helpful in diagnosing to the maximum of seven movements. Acceleration and angular velocity parameters can be used to extract the attributes that are relevant to time and frequency domain. These attributes are then offered to a classification algorithm. An attempt has been made to check the outcomes of four different procedures for categorizing manual movements. The above-mentioned procedures are the Artificial Neural Network (ANN), K-Nearest Neighbours (KNN), Quadratic Support Vector Machine (QSVM), and FP growth. Power spectral density of the acceleration is used to maximize the outcomes of the classifier. Acceleration data alone is considered for activity discovery in the initial step. Experimental results show that the KNN, ANN, QSVM, and FP growth procedures can attain correctness of 81.2%, 87.8%, 93.2%, and 94.1%. The accuracy of fall detection touches 97.2% and 99.1% with no false positive values for the QSVM and FP growth procedures. As a next phase, the attributes can be taken autocorrelation function and the power spectral density of the acceleration and the angular velocity data that obviously maximizes the classification correctness. Projected attributes are adopted to achieve notable improvement in accuracy of 85.8%, 91.8%, 96.1%, and 97.7% for the KNN, ANN, QSVM, and FP growth algorithm. The accurateness of fall discovery touches 100% for both the QSVM and FP growth procedures in the absence of any incorrect alarm, which is the finest attainable performance.

Automated major depressive disorder detection using melamine pattern with EEG signals

Major depressive disorder (MDD) is one of the most common modern ailments affected huge population throughout the world. The electroencephalogram (EEG) signal is widely used to screen the MDD. The manual diagnosis of MDD using EEG is time consuming, subjective and may cause human errors. Therefore, nowadays various automated systems have been developed to diagnose MDD accurately and rapidly. In this work, we have proposed a novel automated MDD detection system using EEG signals. Our proposed model has three steps: (i) Melamine pattern and discrete wavelet transform (DWT)- based multileveled feature generation, (ii) selection of most relevant features using neighborhood component analysis (NCA) and (iii) classification using support vector machine (SVM) and k nearest neighbor (kNN) classifiers. The novelty of this work is the application of melamine pattern. The molecular structure of melamine (also named chemistry spider- ChemSpider) is used to generate 1536 features. Also, various statistical features are extracted from DWT coefficients. The NCA is used to select the most relevant features and these selected features are classified using SVM and kNN classifiers. The presented model attained greater than 95% accuracies using all channels with quadratic SVM classifier. Our results obtained highest classification accuracy of 99.11% and 99.05% using Weighted kNN and Quadratic SVM respectively using A2A1 EEG channel. We have developed the automated depression model using a big dataset and yielded high classification accuracies. These results indicate that our presented model can be used in mental health clinics to confirm the manual diagnosis of psychiatrists.

Computer-vision classification of corn seed varieties using deep convolutional neural network

Automated classification of seed varieties is of paramount importance for seed producers to maintain the purity of a variety and crop yield. Traditional approaches based on computer vision and simple feature extraction could not guarantee high accuracy classification. This paper presents a new approach using a deep convolutional neural network (CNN) as a generic feature extractor. The extracted features were classified with artificial neural network (ANN), cubic support vector machine (SVM), quadratic SVM, weighted k-nearest-neighbor (kNN), boosted tree, bagged tree, and linear discriminant analysis (LDA). Models trained with CNN-extracted features demonstrated better classification accuracy of corn seed varieties than models based on only simple features. The CNN-ANN classifier showed the best performance, classifying 2250 test instances in 26.8 s with classification accuracy 98.1%, precision 98.2%, recall 98.1%, and F1-score 98.1%. This study demonstrates that the CNN-ANN classifier is an efficient tool for the intelligent classification of different corn seed varieties.

Detection of prenatal alcohol exposure using machine learning classification of resting-state functional network connectivity data.

Fetal Alcohol Spectrum Disorder (FASD), a wide range of physical and neurobehavioral abnormalities associated with prenatal alcohol exposure (PAE), is recognized as a significant public health concern. Advancements in the diagnosis of FASD have been hindered by a lack of consensus in diagnostic criteria and limited use of objective biomarkers. Previous research from our group utilized resting-state functional magnetic resonance imaging (fMRI) to measure functional network connectivity (FNC), which revealed several sex- and region-dependent alterations in FNC as a result of moderate PAE relative to controls. Considering that FNC is sensitive to moderate PAE, this study explored the use of FNC data and machine learning methods to detect PAE among a sample of rodents exposed to alcohol prenatally and controls. We utilized previously acquired resting state fMRI data collected from adult rats exposed to moderate levels of prenatal alcohol (PAE) or a saccharin control solution (SAC) to assess FNC of resting state networks extracted by spatial group independent component analysis (GICA). FNC data were subjected to binary classification using support vector machine (SVM) -based algorithms and leave-one-out-cross validation (LOOCV) in an aggregated sample of males and females (n=48; 12 male PAE, 12 female PAE, 12 male SAC, 12 female SAC), a males-only sample (n=24; 12 PAE, 12 SAC), and a females-only sample (n=24; 12 PAE, 12 SAC). Results revealed that a quadratic SVM (QSVM) kernel was significantly effective for PAE detection in females. QSVM kernel-based classification resulted in accuracy rates of 62.5% for all animals, 58.3% for males, and 79.2% for females. Additionally, qualitative evaluation of QSVM weights implicates an overarching theme of several hippocampal and cortical networks in contributing to the formation of correct classification decisions by QSVM. Our results suggest that binary classification using QSVM and adult female FNC data is a potential candidate for the translational development of novel and non-invasive techniques for the identification of FASD.

Can Trunk Acceleration Differentiate Stroke Patient Gait Patterns Using Time- and Frequency-Domain Features?

This study classified the gait patterns of normal and stroke participants by using time- and frequency-domain features obtained from data provided by an inertial measurement unit sensor placed on the subject’s lower back (L5). Twenty-three participants were included and divided into two groups: healthy group (young and older adults) and stroke group. Time- and frequency-domain features from an accelerometer were extracted, and a feature selection method comprising statistical analysis and signal-to-noise ratio (SNR) calculation was used to reduce the number of features. The features were then used to train four Support Vector Machine (SVM) kernels, and the results were subsequently compared. The quadratic SVM kernel had the highest accuracy (93.46%), as evaluated through cross-validation. Moreover, when different datasets were used on model testing, both the quadratic and cubic kernels showed the highest accuracy (96.55%). These results demonstrated the effectiveness of this study’s classification method in distinguishing between normal and stroke gait patterns, with only using a single sensor placed on the L5.

A novel approach for designing authentication system using a picture based P300 speller.

Due to great advances in the field of information technology, the need for a more reliable authentication system has been growing rapidly for protecting the individual or organizational assets from online frauds. In the past, many authentication techniques have been proposed like password and tokens but these techniques suffer from many shortcomings such as offline attacks (guessing) and theft. To overcome these shortcomings, in this paper brain signal based authentication system is proposed. A Brain-Computer Interface (BCI) is a tool that provides direct human-computer interaction by analyzing brain signals. In this study, a person authentication approach that can effectively recognize users by generating unique brain signal features in response to pictures of different objects is presented. This study focuses on a P300 BCI for authentication system design. Also, three classifiers were tested: Quadratic Discriminant Analysis (QDA), K-Nearest Neighbor, and Quadratic Support Vector Machine. The results showed that the proposed visual stimuli with pictures as selection attributes obtained significantly higher classification accuracies (97%) and information transfer rates (37.14 bits/min) as compared to the conventional paradigm. The best performance was observed with the QDA as compare to other classifiers. This method is advantageous for developing brain signal based authentication application as it cannot be forged (like Shoulder surfing) and can still be used for disabled users with a brain in good running condition. The results show that reduced matrix size and modified visual stimulus typically affects the accuracy and communication speed of P300 BCI performance.

Robust classification of neovascularization using random forest classifier via convoluted vascular network

Diabetic Retinopathy is a severe visual disorder in the retina, which causes permanent blindness due to prolonged hyperglycemia. This paper primarily focuses on the classification of Neovascularization into Neovascularization on Disc (NVD) and Neovascularization Elsewhere (NVE). Neovascularization is an alarming phase of Proliferative Diabetic Retinopathy (PDR) that leads to visual impairment of the human eye. This research article proposes a novel method for accurate vessel segmentation using 2-dimensional Gabor Wavelets and matrix convolution. The matrix convolution method is incorporated to preserve the shape of the vascular map and to reduce the impact of unwanted noise on the segmented vascular map. Thirty-six pertinent features are extracted from the vascular map and fed to the Random Forest (RF) classifier as an input for the effective classification of retinal images into Healthy, NVD, and NVE images. The efficacy of the proposed method is estimated on globally accessible databases by calculating the accuracy and Out of Bag (OOB) error. The performance of the RF classifier is compared with Quadratic Support Vector Machine (Q-SVM), Online Sequential Extreme Learning Machine (OS-ELM) and Mediods classifiers. The combination of matrix convolution and feature selection with RF classifier outperforms other existing classifiers by achieving excellent results with an Out of Bag (OOB) error less than 0.05 and an average accuracy of 98 %.

Mapping the Intratumoral Heterogeneity in Glioblastomas with Hyperspectral Stimulated Raman Scattering Microscopy.

Recent genomic studies on the glioblastoma (GBM) subtypes (e.g., mesenchymal, proneural, and classical) pave a way for effective clinical treatments of the recurrent brain tumor. However, identification of the GBM subtype is complicated by the intratumoral heterogeneity that results in coexistence of multiple subtypes within the tissue specimen. Here, we present the use of hyperspectral stimulated Raman scattering (SRS) microscopy for rapid, label-free molecular assessment of GBM intratumoral heterogeneity with submicron resolution. We develop a unique label-free Raman imaging diagnostic platform consisting of the spectral focusing hyperspectral SRS imaging of the large-area GBM tissue specimens, SRS images, and spectrum retrieval using the multivariate curve resolution algorithm and subtype classification based on the quadratic support vector machine model for rapid molecular subtyping of GBMs. Both the stain-free SRS histological images and 2D subtype maps can be obtained within 20-30 min which is superior to the days of the conventional single-cell RNA sequencing. While the SRS histology assesses the demyelination status as a new diagnostic feature, the SRS mapping provides a new insight into intratumoral heterogeneity across GBM tissue specimens. We find that the major proportions of the GBM tissues agree with the diagnostic results of the genomic analysis, but nontrivial portions of the remaining SRS image tiles in the specimens are found to belong to other molecular subtypes, implying the substantial degree of GBM heterogeneity. The rapid SRS imaging diagnostic platform developed has shown the ability of unveiling tumor heterogeneity in GBM tissues accurately, which would promote the improvement of the GBM-targeted therapy in near future.

Evaluation of wavelength ranges and tissue depth probed by diffuse reflectance spectroscopy for colorectal cancer detection

Colorectal cancer (CRC) is the third most common type of cancer worldwide and the second most deadly. Recent research efforts have focused on developing non-invasive techniques for CRC detection. In this study, we evaluated the diagnostic capabilities of diffuse reflectance spectroscopy (DRS) for CRC detection by building 6 classification models based on support vector machines (SVMs). Our dataset consists of 2889 diffuse reflectance spectra collected from freshly excised ex vivo tissues of 47 patients over wavelengths ranging from 350 and 1919 nm with source-detector distances of 630-µm and 2500-µm to probe different depths. Quadratic SVMs were used and performance was evaluated using twofold cross-validation on 10 iterations of randomized training and test sets. We achieved (93.5 ± 2.4)% sensitivity, (94.0 ± 1.7)% specificity AUC by probing the superficial colorectal tissue and (96.1 ± 1.8)% sensitivity, (95.7 ± 0.6)% specificity AUC by sampling deeper tissue layers. To the best of our knowledge, this is the first DRS study to investigate the potential of probing deeper tissue layers using larger SDD probes for CRC detection in the luminal wall. The data analysis showed that using a broader spectrum and longer near-infrared wavelengths can improve the diagnostic accuracy of CRC as well as probing deeper tissue layers.

Microscopic image analysis for herbal plant classification

An identification of herbal plants from its powder form is a challenging task. In this paper, a new method for identification and classification of herbal plants liquorice, rhubarb and dhatura using the microscopic image is proposed. This paper evaluates the effectiveness of the shape and texture-based features with a different classifier for herbal plants classification. Three-shape and five-texture features are computed for each object. The effectiveness of the individual shape and texture-based features set and their combinations are investigated using a support vector machine, K-nearest neighbour and ensemble classifier. The highest 94.9% classification accuracy was achieved by combining all shape features using the bagged tree ensemble classifier. While using a combination of texture-based features almost 99.8% classification accuracy is obtained using fine K-nearest neighbour and cubic-support vector machine classifier. Further, by combining shape and texture-based features classification efficiency achieved is 99.3% with quadratic-support vector machine. From the analysis of simulation results, it is found that texture-based features are more effective to classify a microscopic image of herbal plants.

An Event-Triggered Low-Cost Tactile Perception System for Social Robot’s Whole Body Interaction

The social interaction is one of the necessary skills for social robots to better integrate into human society. However, current social robots interact mainly through audio and visual means with little reliance on haptic interaction. There still exist many obstacles for social robots to interact through touch: 1) the complex manufacturing process of the tactile sensor array is the main obstacle to lowering the cost of production; 2) the haptic interaction mode is complex and diverse. There are no social robot interaction standards and data sets for tactile interactive behavior in the public domain. In view of this, our research looks into the following aspects of tactile perception system: 1) Development of low-cost tactile sensor array, including sensor principle, simulation, manufacture, front-end electronics, examination, then applied to the social robot’s whole body; 2) Establishment of the tactile interactive model and an event-triggered perception model in a social interactive application for the social robot, then design preprocessing and classification algorithm. In this research, we use k-nearest neighbors, tree, support vector machine and other classification algorithms to classify touch behaviors into six different classes. In particular, the cosine k-nearest neighbors and quadratic support vector machine achieve an overall mean accuracy rate of more than 68%, with an individual accuracy rate of more than 80%. In short, our research provides new directions in achieving low-cost intelligent touch interaction for social robots in a real environment. The low-cost tactile sensor array solution and interactive models are expected to be applied to social robots on a large scale.

A family of divergence-based classifiers for Polarimetric Synthetic Aperture Radar (PolSAR) imagery vector and matrix features

ABSTRACT Polarimetric Synthetic Aperture Radar (PolSAR) is one of the most important remote sensing tools. However, PolSAR images are strongly contaminated by a multidimensional interference (called speckle noise), making their processing (e.g. in the classification context) difficult. In terms of structure, multilook PolSAR data follow a definite positive hermitian behaviour and, therefore, require tailored classifiers for their features. Some classic classifiers – such as Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), K-Nearest Neighbours (KNN), and Support Vector Machine (SVM) – have been yielding unacceptable performance to these data, when applied directly. One justification is because they do not often take into account neither the speckle presence nor properties which are inherent to under-study relieves. This paper addresses a collection of PolSAR divergence-based classifiers, deduced from the normal, skew-normal, t-Student, and skew-t vector models as well as the scaled complex Wishart (SCW) distribution. The last model is a standard supposition to describe multilook PolSAR data, having two parameters: covariance matrix (which is directed to data nature) and number of looks (which controls the speckle noise effect). The considered remainder laws aim to model the main diagonal of these data, known as multivariate intensities. The performance of proposed methods is quantified and compared with those due to the Kullback-Leibler (KL) distance for multivariate normal distribution and to LDA, QDA, KNN, and SVM methods. Experiments with both artificial and real PolSAR data are considered. Results favour optimal Rényi classifiers for an Airborne Synthetic Aperture Radar (AIRSAR) image of San Francisco and the t-Student KL classifier for an SAR image system of the Electromagnetics Institute (EMISAR) image of Foulum.

Predictive modeling of proliferative vitreoretinopathy using automated machine learning by ophthalmologists without coding experience

We aimed to assess the feasibility of machine learning (ML) algorithm design to predict proliferative vitreoretinopathy (PVR) by ophthalmologists without coding experience using automated ML (AutoML). The study was a retrospective cohort study of 506 eyes who underwent pars plana vitrectomy for rhegmatogenous retinal detachment (RRD) by a single surgeon at a tertiary-care hospital between 2012 and 2019. Two ophthalmologists without coding experience used an interactive application in MATLAB to build and evaluate ML algorithms for the prediction of postoperative PVR using clinical data from the electronic health records. The clinical features associated with postoperative PVR were determined by univariate feature selection. The area under the curve (AUC) for predicting postoperative PVR was better for models that included pre-existing PVR as an input. The quadratic support vector machine (SVM) model built using all selected clinical features had an AUC of 0.90, a sensitivity of 63.0%, and a specificity of 97.8%. An optimized Naïve Bayes algorithm that did not include pre-existing PVR as an input feature had an AUC of 0.81, a sensitivity of 54.3%, and a specificity of 92.4%. In conclusion, the development of ML models for the prediction of PVR by ophthalmologists without coding experience is feasible. Input from a data scientist might still be needed to tackle class imbalance—a common challenge in ML classification using real-world clinical data.

Classification of working memory loads using hybrid EEG and fNIRS in machine learning paradigm

Single modality brain–computer interface (BCI) systems often mislabel the electroencephalography (EEG) signs as a command, even though the participant is not executing some task. In this Letter, the classification of different working memory load levels is presented using a hybrid BCI system. N-back cognitive tasks such as 0-back, 2-back, and 3-back are used to create working memory load on participants while recording EEG and functional near-infrared spectroscopy (fNIRS) signals simultaneously. A combination of statistically significant features obtained from EEG and fNIRS corresponding to frontal region channels are used to classify different N-back commands. Kernel-based support vector machine (SVM) classifiers are employed with and without cross-validation schemes. Classification accuracy of 100% is achieved for binary classification of 0-back against 2-back and 0-back against 3-back using linear SVM, quadratic SVM, and cubic SVM under holdout data division protocol.

A kernel-free double well potential support vector machine with applications

As a well-known machine learning technique, support vector machine (SVM) with a kernel function achieves much success in nonlinear binary classification tasks. Recently, some quadratic surface SVM models are proposed and studied by utilizing quadratic surfaces for nonlinear binary separations. In this paper, a kernel-free soft quartic surface SVM model is proposed by utilizing the double well potential function for highly nonlinear binary classification. Mathematical analysis on the theoretical properties of the proposed model, including the existence, uniqueness and support vector representation of optimal solutions, is shown. The sequential minimal optimization algorithm is adopted to implement the proposed model for computational efficiency. Numerical results on some artificial and public benchmark data sets demonstrate its effectiveness over well-known SVM models with or without kernel functions. The proposed model is extended to successfully handle some real-life corporate and personal credit data sets for applications.

Authentication framework for security application developed using a pictorial P300 speller

ABSTRACT With the recent advancement in the field of information technology, online frauds have also been increased and hence password-based systems are no more secure. Therefore need for a more reliable authentication system has been growing rapidly. This study focuses on a P300 based authentication system designing. The proposed study presents a new four-choice oddball paradigm where a conventional 6 × 6 character-based speller is replaced by a novel 2 × 2 pictorial speller. In this study, three classifiers were tested: Quadratic Discriminant Analysis (QDA), K-Nearest Neighbor (KNN), and Quadratic Support Vector Machine (QSVM). Out of these three, QDA performed better. The results showed that the designed visual stimuli obtained significantly higher accuracy (97%) and information transfer rates (37.14 bits/min) as compared to the conventional paradigm. The results also show that reduced matrix size and modified stimulus typically affects the accuracy and communication speed of a P300 based authentication system.

GC-MS-based untargeted metabolomics of plasma and urine to evaluate metabolic changes in prostate cancer

Prostate cancer (CaP) is a common cancer in men. Its late detection and inefficient diagnosis are a challenge for researchers who are currently searching for new cancer-related indicators that would facilitate better detectability of CaP and explain its pathogenesis. In the present preliminary study, endogenous volatile metabolites were detected in plasma and urine samples by using the metabolic fingerprinting approach. The analyses were performed using the GC-QqQ/MS technique in the scan mode. The detected and putatively identified metabolites were statistically analyzed using advanced univariate and multivariate statistical methods. Eleven urinary and three plasma metabolites were selected as statistically significant in patients with CaP as compared to those in healthy controls. Supervised methods such as logistic regression and quadratic support vector machine were applied to obtain the classification models. The accuracy, sensitivity, and specificity of the models were above 83%, 85%, and 81%, respectively. The putatively identified metabolites were associated with biochemical pathways such as tricarboxylic acid cycle, glycolysis, carbohydrate conversion, and steroidal lipid metabolism that are mainly involved in energy production for cell growth and proliferation.

Fall Detection and Daily Living Activity Recognition using Machine Learning

Elder people are increasing all over the world as a result certain fall occur in their daily life. This fall lead to several severe problems. The fall may often causes injuries and in many cases it result in death of the individual. The problem should be addressed to reduce the fall. By using some Machine Learning(ML) algorithm the fall and daily living activities are recognized. The acceleration and angular velocity data obtained from the dataset are used to detect the fall and daily living activity. Body movement of the person are collected and stored in the dataset. Acceleration and angular velocity data are used to extract the time and frequency domain feature and provide them to classification algorithm. Here, Logistic regression algorithm is used for detecting the fall and living activity. It is very effective algorithm and does not require too many computational resources. It is easy to regularize and provide well calibrated predicted probabilities as output. The sensitivity, accuracy and specificity of fall detection and activity recognition is obtained as a result. The performance evaluation is made with three classification algorithm. The three classification algorithm are Artificial neural network (ANN), K-nearest neighbours (KNN), Quadratic support vector machine (QSVM). Logistic regression provides highest accuracy compared with other three algorithm.

Comparative Study of Machine Learning Algorithms to Classify Hand Gestures from Deployable and Breathable Kirigami-Based Electrical Impedance Bracelet

Wearable devices are gaining recognition for their use as a biosensor platform. Electrical impedance tomography (EIT) is one of the sensing techniques that utilizes wearable sensors as its primary data acquisition system. It measures the impedance or resistance at the peripheral (skin) level and calculates the conductivity distribution throughout the body. Even though the technology has existed for several decades, modern-day EIT devices are still costly and bulky. The paper proposes a novel low-cost kirigami-based wearable device that has soft PEDOT: PSS electrodes for sensing skin impedances. Simulation results show that the proposed kirigami structure for the bracelet has a large deformation during actuation while experiencing relatively lower stress. The paper also presents a comparative study on a few machine learning algorithms to classify hand gestures, based on the measured skin impedance. The best classification accuracy (91.49%) was observed from the quadratic support vector machine (SVM) algorithm with 48 principal components.

Extracted features based multi-class classification of orthodontic images

The purpose of this study is to investigate computer vision and machine learning methods for classification of orthodontic images in order to provide orthodontists with a solution for multi-class classification of patients’ images to evaluate the evolution of their treatment. Of which, we proposed three algorithms based on extracted features, such as facial features and skin colour using YCbCrcolour space, assigned to nodes of a decision tree to classify orthodontic images: an algorithm for intra-oral images, an algorithm for mould images and an algorithm for extra-oral images. Then, we compared our method by implementing the Local Binary Pattern (LBP) algorithm to extract textural features from images. After that, we applied the principal component analysis (PCA) algorithm to optimize the redundant parameters in order to classify LBP features with six classifiers; Quadratic Support Vector Machine (SVM), Cubic SVM, Radial Basis Function SVM, Cosine K-Nearest Neighbours (KNN), Euclidian KNN, and Linear Discriminant Analysis (LDA). The presented algorithms have been evaluated on a dataset of images of 98 different patients, and experimental results demonstrate the good performances of our proposed method with a high accuracy compared with machine learning algorithms. Where LDA classifier achieves an accuracy of 84.5%.