Classification Results Of SVM Research Articles

Characterization of Bimanual Cyclical Tasks From Single-Trial EEG-fNIRS Measurements.

Robot-assisted bimanual training is promising to improve motor function and cortical reorganization for hemiparetic stroke patients. Closing the rehabilitation training loop with neurofeedback can help refine training protocols in time for better engagements and outcomes. However, due to the low signal-to-noise ratio (SNR) and non-stationary properties of neural signals, reliable characterization of bimanual training-induced neural activities from single-trial measurement is challenging. In this study, ten human participants were recruited conducting robot-assisted bimanual cyclical tasks (in-phase, 90° out-of-phase, and anti-phase) when concurrent electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) were recorded. A unified EEG-fNIRS bimodal signal processing framework was proposed to characterize neural activities induced by three types of bimanual cyclical tasks. In this framework, novel artifact removal methods were used to improve the SNR and the task-related component analysis (TRCA) was introduced to increase the reproducibility of EEG-fNIRS bimodal features. The optimized features were transformed into low-dimensional indicators to reliably characterize bimanual training-induced neural activation. The SVM classification results of three bimanual cyclical tasks revealed a good discrimination ability of EEG-fNIRS bimodal indicators (90.1%), which was higher than that using EEG (74.8%) or fNIRS (82.2%) alone, supporting the proposed method as a feasible technique to characterize neural activities during robot-assisted bimanual training.

A novel method for classification of multi-class motor imagery tasks based on feature fusion

Motor imagery based brain computer interface (MI-BCI) has the advantage of strong independence that can rely on the spontaneous brain activity of the user to operate external devices. However, MI-BCI still has the problem of poor control effect, which requires more effective feature extraction algorithms and classification methods to extract distinctly separable features from electroencephalogram (EEG) signals. This paper proposes a novel framework based on Bispectrum, Entropy and common spatial pattern (BECSP). Here we use three methods of bispectrum in higher order spectra, entropy and CSP to extract MI-EEG signal features, and then select the most contributing features through tree-based feature selection algorithm. By comparing the classification results of SVM, Random Forest, Naive Bayes, LDA, KNN, Xgboost and Adaboost, we finally decide to use the SVM algorithm based on RBF kernel function which obtained the best result among them for classification. The proposed method is applied to the BCI competition IV data set 2a and BCI competition III data set IVa. On data set 2a, the highest accuracy on the evaluation data set reaches 85%. The experiment on data set IVa can also achieve good results. Compared with other algorithms that use the same data set, the performance of our algorithm has also been improved.

Attack sample generation algorithm based on data association group by GAN in industrial control dataset

The importance of industrial control networks security is growing, but the intrusion detection research of industrial control networks is seriously restricted by the existing attack samples of the business dataset, especially the quantity and quality. In order to solve the problem of the scarcity of attack industrial control datasets, this paper proposes an attack sample generation algorithm. Firstly, based on the weight and degree of membership distribution, calculate the value of membership distance between dimensions, and the data association is strong when the membership distance of dimensions is small. Then, divide dimensions which have small distance into a group, so as to realize the association grouping of the original data. The data association of dimensions in an association group is strong when the association group appears frequently. According to the frequency of the association group, all the association groups are divided into strong association group and weak association group. Attack all the dimensions of one strong association group in the original data by false data injection attack, realized attack sample generation algorithm in the original data. Finally, expand the attack sample into a large amount of attack sample industrial control dataset by the Generative Adversarial Network. In this paper, the attack samples are generated by the BATADAL dataset and the business dataset of an oil depot, and the data is expanded by 100 times through the algorithm. Compared with the attack samples provided by the BATADAL dataset, the coincidence degree and fitting degree of generated data is improved by 38.20%–42.94% and 98.22%–98.36%, respectively. The classification results of XGBoost and SVM are 100% and 98.01%, which is close to the classification result of attack samples provided by BATADAL dataset.

The building recognition and analysis of remote sensing image based on depth belief network

The deep belief network model, which is widely used in deep learning, consists of a multi-layer constrained Boltzmann machine and a back-propagation network. The authors have conducted parameter sensitivity experiments on the number of iterations, the number of hidden layers and the number of hidden layer nodes in the DBN network for remote sensing image classification, and obtained a set of optimal parameter setting schemes. Moreover, the DBN algorithm has been enhanced with an improved Dropout strategy. The improved Dropout strategy selects only part of the data to clear the weight at a time, and a local area randomly clear strategy is adopted, which will save the local information of the image itself, and enhance the generalization ability of the model. In order to verify the advantages of the improved DBN algorithm model, the classification results of DBN, KNN, random forest and SVM have been compared. And the results show that classification accuracy of the improved DBN has been greatly improved, which is increased by about 2.5% compared to DBN. The improved DBN classification results are processed then, including connected areas marking, noise removal, morphological transformation and edge extraction, and the boundary information of the building is obtained according to the target shape characteristics. Finally, the experiment on the morphological characteristics of the building also shows it can extract better edge information of the building.

Prediction of Damage to Insulation Joints Based on SVM with Unbalanced Data Sets

As a key part of track circuit, the state of insulation joints is related to safe, normal and efficient operation of railway. In order to accurately obtain different degrees of insulation joints, a prediction model based on support vector machines has been proposed to study damage to insulation joints. For unbalanced data sets in the research process, a KNN under-sampling is presented to remove redundant and noise samples. By means of BSMOTE over-sampling method to further take full advantage of the data, KNN-BSMOTE-SVM algorithm of hybrid sampling is given to achieve balanced data sets. The theoretical analysis and simulation results show that the proposed algorithm increases classification performance of SVM classifier. Compared with KNN classifier, the classification results of SVM are better, support vector machines used in insulation damaged joints prediction is feasible and effective.

An Empirical Analysis of Imbalanced Data Classification

SVM has been given top consideration for addressing the challenging problem of data imbalance learning. Here,we conduct an empirical classification analysis of new UCI datasets that have dierent imbalance ratios, sizes andcomplexities. The experimentation consists of comparing the classification results of SVM with two other popularclassifiers, Naive Bayes and decision tree C4.5, to explore their pros and cons. To make the comparative exper-iments more comprehensive and have a better idea about the learning performance of each classifier, we employin total four performance metrics: Sensitive, Specificity, G-means and time-based eciency. For each benchmarkdataset, we perform an empirical search of the learning model through numerous training of the three classifiersunder dierent parameter settings and performance measurements. This paper exposes the most significant resultsi.e. the highest performance achieved by each classifier for each dataset. In summary, SVM outperforms the othertwo classifiers in terms of Sensitive (or Specificity) for all the datasets, and is more accurate in terms of G-meanswhen classifying large datasets.

Comparison of the efficacy of spectral pre-treatments for wheat and weed discrimination in outdoor conditions

The optimal processing of spectral data often requires specific pre-treatments. In the context of spectral discrimination, results can be greatly improved using the relevant pre-treatment. Most importantly, the pre-treatment must be suited to the nuisance variability that has to be removed. This study focuses on discrimination of weed and wheat using spectra acquired in outdoor conditions with uncontrolled lighting and leaf orientations. Both vegetation spectra are highly similar, and due to the context of acquisition, a lot of spectral variability is present. This nuisance variability is modeled using an additive, multiplicative and noise term, each of which affects the measured spectra. Several pre-treatments were therefore evaluated according to their potential to deal with this variability and their effects were described in the feature space. Finally, results obtained with these pre-treatments combined with two discrimination methods (PLS-LDA and Gaussian SVM) are compared and discussed. Results showed that, thanks to their ability to remove nuisance variability, most pre-treatments are effective in terms of classification accuracy. Gaussian SVM classification results are less influenced by pre-treatments than those of PLS-LDA, since the former compensates the pre-treatment effect by using a different non-linear kernel. For this data-set, the best discrimination result was obtained using the combination logarithm and PLS-LDA. Logarithm actually transforms the multiplicative effect into and additive one, which is then effectively dealt with by PLS-LDA.

An Improved Method of SVM-BPSO Feature Selection Based on Cloud Model

An improved method of SVM-BPSO feature selection based on cloud model is proposed to solve the local deadlock problem of the current SVM-BPSO methods. This method uses the Wrapper evaluation strategy, making SVM as the classifier, BPSO algorithm is adopted to conduct a whole search in the feature space to seek out the optimal feature subset from the classification results of SVM. The inertia weight of BPSO algorithm is adjusted by Cloud Model intelligently and self-adaptively, the whole and local searching capability of SVM-BPSO feature selection algorithm get balanced, and prevent it into the local deadlock effectively. Analysis on simulations shows that the improved method of SVM-BPSO feature selection based on Cloud Model can be able to dap out from the local optimum with a faster convergence rate, and shows good experimental effect. DOI : http://dx.doi.org/10.11591/telkomnika.v12i5.5005

Feature Extension for Short Text Categorization Using Frequent Term Sets

A short text feature extension method based on frequent term sets is proposed to overcome the drawbacks of the vector space model (VSM) on representing short text content. After defining the co-occurring and class orientation relations between terms, frequent term sets with identical class orientation are generated by calculating the support and confidence of word sets, and then taken as the background knowledge for short text feature extension. For each single term of the short text, the term sets containing this term are retrieved in the background knowledge and added into the original term vector as the feature extension. The experimental results on Sougou corpus show that the support and confidence have great impact on the scale of the background knowledge, but excessive extension also has redundancy and cannot obtain further improvement. The background knowledge based on frequent term sets is an effective way for feature extension. When the number of the training documents is limited, these extended features can greatly improve the classification results of SVM.