Linear and nonlinear speech feature analysis for stress classification

EEG signals are non-stationary, complex and non-linear signals. During major depressive disorder (MDD) or depression, any deterioration in the brain function is reflected in the EEG signals. In this paper, linear features (band power, inter hemispheric asymmetry) and non-linear features [relative wavelet energy (RWE) and wavelet entropy (WE)] and combination of linear and non-linear features were used to classify depression patients and healthy individuals. In this analysis the data set used is publicly available data set contributed by Mumtaz et al. (Biomed Signal Process Control 31:108–115, 2017b). The dataset consisted of 34 MDD patients and 30 healthy individuals. The classifiers used were multi layered perceptron neural network (MLPNN), radial basis function network (RBFN), linear discriminant analysis (LDA) and quadratic discriminant analysis. When linear feature was used, highest classification accuracy of 91.67% was obtained by alpha power with MLPNN classifier. When non-linear feature was used, both RWE and WE provided highest classification accuracy of 90% with RBFN and LDA classifier, respectively. The highest classification of 93.33% was achieved when combining linear and non-linear feature, i.e., combination alpha power and RWE with MLPNN as well as RBFN classifier. This paper also showed that the combination of non-linear features, i.e., RWE and WE also performed the best with highest classification accuracy of 93.33%. The study compared the accuracy, sensitivity and specificity of different classifiers along with linear and non-linear features and combination of both. The results indicated that combination alpha power and RWE showed the highest classification 93.33% accuracy in all the applied classifiers.

Three systems for the classification of stress in speech are proposed. The first system makes use of linear short time log frequency power coefficients (LFPC), the second employs a Teager energy operator (TEO) based nonlinear frequency domain LFPC features (NFD-LFPC) and the third uses TEO based nonlinear time domain LFPC features (NTD-LFPC). The systems were tested using the SUSAS (speech under simulated and actual stress) database to categorize five stress conditions individually. Results show that the system using LFPC gives the highest accuracy, followed by the system using NFD-LFPC features, while the system using NTD-LFPC features gives the worst performance. For the system using linear LFPC features, average accuracy of 84% and best accuracy of 95% were obtained in classifying five stress categories.

Microbes play a crucial role in human health and disease. Understanding the relationship between microbes and diseases is conducive to the treatment and diagnosis of diseases. Recently, many computational methods have been proposed to predict disease-microbe associations. However, most of the existing methods only consider a single model and explore the disease-microbe associations from a single view. To improve the prediction accuracy, we propose a novel multi-view approach that fuses the linear and nonlinear features to predict new potential associations between diseases and microbes. We first design a non-negative matrix tri-factorization method to extract the linear features of diseases and microbes. We input the linear features from the non-negative matrix tri-factorization model and bi-random walk model into a capsule neural network to obtain the diseases and microbes’ nonlinear features. Finally, we leverage a logistic regression model to combine the non-negative matrix tri-factorization model predictions, bi-random walk model predictions and the capsule neural network predictions to obtain the final association scores between microbes and diseases. We apply our method to predict human microbedisease associations. Experimental results show that our fusion model outperforms the non-negative matrix tri-factorization model, bi-random walk model and other existing models.

Electrocardiogram analysis using a combination of statistical, geometric, and nonlinear heart rate variability features

Insomnia is a sleep disorder that causes disturbance in a normal sleep pattern resulting from the difficulties to fall asleep or to stay asleep. Biosignals of human body if measured are also affected due to these abnormal conditions. A current practise in diagnosing insomnia is through clinical interview by the physician which is subjective and suffers from human error judgement. Therefore, a more reliable and accurate diagnostic tools are needed to help physician in making decision. The objective of this study is to classify healthy and insomnia by implementing advanced classification technique based on Artificial Neural Network (ANN).In this study, sleep EEG and ECG signals of 10 insomnia patients and 10 healthy subjects are analysed. Several linear and nonlinear features are extracted from the denoised signals: linear features (power spectral of EEG frequency bands, brain rate, Hjorth parameters, heart rate variability) and nonlinear features (Largest Lyapunov Exponent (LLE), Sample Entropy and Correlation Dimension (CD)).For classification purpose, a Feedforward Neural Network (FNN) is implemented to classify the two groups. Half of the data is used for training and the other half is used for testing the classifier. The Levenberg Marquardt backpropagation algorithm is used as a training function. Several numbers of hidden layers are tested in order to achieve optimum classification accuracy. A classification accuracy of 81.3 % is obtained for 3 hidden layers This result shows that the combination of features extracted from EEG and ECG signals during sleep and FFN are useful to be adopted as a biomarker and classifier in identifying insomnia.of FNN.

Considering the non-stationary characteristics of Electroencephalogram (EEG) signals has a valuable impact in improving the classification rate for motor imagery tasks recognition. This paper aims to construct an efficient scheme for the classification of right and left hand movement. The scheme is based on combining linear and non-linear features in order to enhance the classification rate. Linear features are extracted from the amplitude frequency analysis (AFA) of the EEG signal. While the non-linear features are extractedfrom a density matrix which is generated from the phase space of the signal. We have used four classification approaches in this study; the linear discriminant analysis (LDA), support vector machine (SVM), Bayes and KNN classifiers. The Graz 2003 datasets has been used in this study. The maximal classification rate we have achieved is 89.3%. Results confirmed the robustness of the new technique and demonstrate its value as a classification approach in the field of brain computer interface BCI.

A new short-term wind power prediction methodology based on linear and nonlinear hybrid models

Electroencephalography (EEG) enables online monitoring brain activity, which can be used for neurofeedback. One of the growing applications of EEG neurofeedback is to facilitate meditation practice. Specifically, EEG neurofeedback can be used to alert participants whenever they get distracted during meditation practice based on changes in their brain activity. In this study, we develop machine learning models to detect moments of distraction (due to mind wandering or drowsiness) during meditation practice using EEG signals. We use EEG data of 24 participants while performing a breath focus meditation with experience sampling and extract twelve linear and nonlinear EEG features. Features are fed to ten supervised machine learning models to classify (i) Breath Focus Awake (BFA) vs Breath Focus Sleepy (BFS), and (ii) BFA vs Mind Wandering (MW). We observe that the linear features achieve a maximum accuracy of 86% for classifying awake (BFA) and sleepy (BFS), whereas non-linear features have more predictive ability for classifying between BFA and MW with a maximum accuracy of nearly 78%. In addition, visualization of unsupervised t-SNE lower embeddings supports the evidence of distinct clusters for each condition. Overall our results show that machine learning algorithms can successfully identify periods of distraction during meditation practice in novice meditators based on linear and non-linear features of the EEG signal. Consequently, our results have important implications for the development of mobile EEG neurofeedback protocols aimed at facilitating meditation practice.

EEG-based mild depressive detection using feature selection methods and classifiers

Epilepsy is a neurological disorder that causes seizures of many different types. Recent research has shown that epileptic seizures can be predicted by using the electrocardiogrami instead of the electroencephalogram. In this study, we used the heart rate variability that is generated by the fluctuating balance of sympathetic and parasympathetic nervous systems to predict epileptic seizures. We studied 11 epilepsy patients to predict the seizure interval. With regar tos the fact that HRV signals are nonstationary, our analysis focused on linear features in the time and frequency domain of HRV signal such as RR Interval (RRI), mean heart rate (HR), high-frequency (HF) (0.15-0.40 Hz) and low-frequency (LF) (0.04-0.15 Hz), as well as LF/HF. Also, quantitative analyses of Poincaré plot features (SD1, SD2, and SD1/SD2 ratio) were performed. HRV signal was divided into intervals of 5 minutes. In each segment linear and nonlinear features were extracted and then the amount of each segment compared to the previous segment using a threshold. Finally, we evaluated the performance of our method using specificity and sensitivity. During seizures, mean HR, LF/HF, and SD2/SD1 ratio significantly increased while RRI significantly decreased. Significant differences between two groups were identified for several HRV features. Therefore, these parameters can be used as a useful feature to discriminate a seizure from a non-seizure The seizure prediction algorithm proposed based on HRV achieved 88.3% sensitivity and 86.2 % specificity. These results indicate that the HRV signal contains valuable information and can be a predictor for epilepsy seizure. Although our results in comparison with EEG ares a little bit weaker, the recording of ECG is much easier and faster than EEG. Also, our finding showed the results of this study are considerably better than recent research based on ECG (Tab. 1, Fig. 10, Ref. 17).

Chapter Three - Role of Different Features in Diagnosis of Alzheimer Disease

The method of geo-referencing satellite- and aerial imagery using reference points, linear and non-linear features, and segments of geodetic tracks as elements of a plan-altitude basis is discussed in this article. The method can be used for any mathematical model of satellite- and aerial imagery. The parametric description of the features, that can be used for the geo-referencing and various mathematical models of the above-mentioned imagery are presented. The mathematical formulation of the matter of satellite and aerial imagery geo-referencing using terrain objects and reference points is presented. A list of linear and non-linear features that can be included in a high-raised basis along with reference points is made. A generalized algorithm for geo-referencing satellite and aerial imagery using reference points and terrain features is given. The algorithm includes making a nonlinear system of equations for reference points and items, linearizing the system and solving by the sequential approximation technique according to the least squares method. An example of clarifying the satellite RPC-model and aerial imagery using reference points, linear and non-linear features is given. The advantages of the proposed method of using features created according to the measurements on satellite and aerial imagery compared with method, when the model of the feature is created according to the measurements on the ground are described.

To study effective speech features which can represent different emotion styles in mandarin speech, nonlinear features based on Teager Energy Operator(TEO) are researched. Neutral state and 3 emotional states (i.e. happiness, anger and sadness) are classified from the mandarin speech database. MFCC extraction and HMM-based emotion recognition are used as baseline system to evaluate the emotional classification performance of TEO-based features. In comparison with MFCC, while text- dependent, improvements of classification capacity are obtained when using all 4 nonlinear features (i.e. NFD_Mel, AF_Mel, DAF_Mel, AM_SBCC). While text-independent, the performance of emotion classification are improved by using NFD_Mel, AF_Mel and DAF_Mel, but deteriorated by using AM_SBCC. The results of classification demonstrate that the nonlinear features based on TEO, when using NFD_Mel, AF_Mel and DAF_Mel, are better able to represent different emotion styles in speech than that of MFCC.

VAE-GANMDA: A microbe-drug association prediction model integrating variational autoencoders and generative adversarial networks.

In this paper, we propose a new method to extract thin linear decorrelated features in SAR interferometric images by fusioning the information provided by the amplitude and the coherence. A first detection is the result of an unsupervised classification performed on the coherence image, where linear features correspond to dark areas (low coherence) and non-linear features to brighter areas. This approximate location of the linear features is further refined by using edge information extracted in the amplitude data by two measures: the coefficient of variation (CV) and the ratio of local means (RLM). A precise detection is then performed by using the results of coherence classification and the fusion of the two previous measures. The method has been applied on ERS SAR images from the western part of Cameroon to extract thin river networks in mangrove areas. The goal is to update - or create geographical maps.