Among the barriers to establishing effective human-machine interactions is the machines' inability to properly distinguish emotions from the human voice. The Speech Emotion Recognition (SER) systems have emerged t...
详细信息
Among the barriers to establishing effective human-machine interactions is the machines' inability to properly distinguish emotions from the human voice. The Speech Emotion Recognition (SER) systems have emerged to tackle this limitation. The accuracy of these systems depends on different factors such as the quantity and the types of emotions included in the database, feature extraction process including local and global features, feature selection method, and the type of classifier. This study presents a methodology for speech emotion recognition using an autoencoder neural network. It is shown that using a digit-level stacked autoencoder can be suitable for digit classification. The speech emotion recognition is done using the Persian emotional speech database (Persian ESD), which includes six emotional states: Happiness, Sadness, Fear, Disgust, Anger, and Neutral. Moreover, the popular, widely-used Berlin Emotional database (EMO-DB) is used to evaluate the effectiveness of the proposed approach. The experimental results show that the proposed method has significantly improved recognition accuracy.
In recent years, several fraud attempts have been made in various sectors including finance, banking and insurance. In fact, credit card fraud refers to the unauthorized use of a credit card account to obtain money, p...
详细信息
In recent years, several fraud attempts have been made in various sectors including finance, banking and insurance. In fact, credit card fraud refers to the unauthorized use of a credit card account to obtain money, products or services. It involves the manipulation of card details for fraudulent purchases or withdrawals. These fraudulent incidents result in substantial financial losses of different divisions of businesses. The present manuscript presents an innovative system used to detect credit card fraud employing unsupervised deep learning. However, the effectiveness of deep learning models relies on the configuration of hyperparameter values and the avoiding of overffiting issue, tasks that provide time-consuming and require significant trial and error. The proposed model utilizes an improved particle swarm optimization (PSO) to optimize the training hyperparameters such as global initial connection weights and thresholds, while leveraging stacked autoencoder for classification purposes. Unlike the existing model, that introduced in this work takes into account transactional data and enables the classifier to accurately identify the most crucial transactions within the input sequence, which allows predicting fraudulent transactions more accurately. This approach combines the strengths of three methods: SMOTE-Tomek for handling imbalanced data, local search PSO for hyperparameter optimization, and an enhanced stacked autoencoder used to detect anomaly in credit card fraud transactions. The comparative study reveals that the developed model is the most efficient, in comparison with the other models.
Most labor contract evaluations rely on performance evaluations by human resource management, which is time-consuming and costly. However, there has been little research into quantitative contract evaluations. This pa...
详细信息
Most labor contract evaluations rely on performance evaluations by human resource management, which is time-consuming and costly. However, there has been little research into quantitative contract evaluations. This paper embedded a stacked autoencoder into a weighted two-stage data envelopment analysis model to evaluate NBA rookie seasonal contracts in an attempt to quantitatively assess contract execution efficiency. It was found that the model was able to effectively evaluate the NBA rookie contracts and provide guidance to the coach regarding their on-court performances. The NBA rookie contract execution analyses also found that performance and therefore contract fulfilment was possibly affected by time allocation problems. Finally, a dynamic and comprehensive contract evaluation system that has significant possible commercial value was constructed to assist the player, coach and manager make timely decisions, which may be a breakthrough in objective human resource management performance evaluation systems.
Trust-aware routing is the significant direction for designing the secure routing protocol in Wireless Sensor Network (WSN). However, the trust-aware routing mechanism is implemented to evaluate the trustworthiness of...
详细信息
Trust-aware routing is the significant direction for designing the secure routing protocol in Wireless Sensor Network (WSN). However, the trust-aware routing mechanism is implemented to evaluate the trustworthiness of the neighboring nodes based on the set of trust factors. Various trust-aware routing protocols are developed to route the data with minimum delay, but detecting the route with good quality poses a challenging issue in the research community. Therefore, an effective method named Sunflower Sine Cosine (SFSC)-based stacked autoencoder is designed to perform Electroencephalogram (EEG) signal classification using trust-aware routing in WSN. Moreover, the proposed SFSC algorithm incorporates Sunflower Optimization (SFO) and Sine Cosine Algorithm (SCA) that reveals an optimal solution, which is the optimal route used to transmit the EEG signal. Initially, the trust factors are computed from the nodes simulated in the network environment, and thereby, the trust-based routing is performed to achieve EEG signal classification. The proposed SFSC-based stacked autoencoder attained better performance by selecting the optimal path based on the fitness parameters, like energy, trust, and distance. The performance of the proposed approach is analyzed using the metrics, such as sensitivity, accuracy, and specificity. The proposed approach acquires 94.708%, 94.431%, and 95.780% sensitivity, accuracy, and specificity, respectively, with 150 nodes.
Accurate traffic-flow prediction is essential to traffic management. Traffic data collected in very short intervals normally contain high variability, while common preprocessing approaches applied within a window such...
详细信息
Accurate traffic-flow prediction is essential to traffic management. Traffic data collected in very short intervals normally contain high variability, while common preprocessing approaches applied within a window such as simple average or median operator are unable to obtain sufficient latent information from original data. Moreover, the prediction performance of shallow neural network is not satisfying, since its capacity to represent the temporal-spatial correlation of mass traffic data is insufficient, and its adaptation capacity to noisy data is relatively poor. In this paper, fuzzy information granulation (FIG) and deep neural network are combined to solve these two issues. To be specific, FIG is utilized to process original data series and extract granules, which denote the trend and fluctuation range of each time window. Then, stacked autoencoder is combined to obtain the predictive results based on processed granules, especially, a multi-output mechanism is designed to predict all granulation parameters simultaneously, which makes better use of the correlation of diverse inputs. A real-world traffic volume data set is applied to conduct an empirical study, and the experimental results illustrate that based on the proposed method, the interval prediction of the traffic-flow fluctuation range is realized, and superior traffic trend prediction performance is achieved.
An Application Layer Distributed Denial of Service Attack (DDoS) is one of the biggest concerns for web security. Many detection methods are designed to mitigate DDoS attack based on IP and TCP layer instead of the Ap...
详细信息
ISBN:
(纸本)9781509000821
An Application Layer Distributed Denial of Service Attack (DDoS) is one of the biggest concerns for web security. Many detection methods are designed to mitigate DDoS attack based on IP and TCP layer instead of the Application layer. These methods are not suitable for detection of Application layer DDoS attack since most of the IP and TCP layer DDoS attacks are based on request flooding attack. But Application layer DDoS attacks consist of request flooding, session flooding, and asymmetric attack. The solutions available to detect Application layer DDoS attack, detect only limited number of Application layer DDoS attacks. The solutions that detect all types of Application layer DDoS attacks have huge algorithm complexity. One of the major challenges in the detection of an Application layer DDoS attack is the non-availability of features to detect such attacks. Hence it is difficult to model normal user behavior from attack behavior. In this paper, Deep learning architecture is introduced to learn deep features of Application layer DDoS attack. Deep learning architecture consist of very deep neural network, typically more than three layers. In the proposed work the concept of autoencoder is applied, which is one of the deep learning based models that learns deep useful features in the Application layer DDoS attack dataset. The stacked autoencoder deep learning architecture, is aimed to receive high level features. The performance of the proposed method was evaluated in terms of the metrics such as false positive rate and detection rate. Comparison of the proposed method with the existing methods reveals that the proposed method performs better than the existing methods.
stacked autoencoders enjoy their popularization with the prosperity of deep learning in recent years. However, relative studies seldom exploit the intrinsic information buried in the interrelations between the samples...
详细信息
ISBN:
(纸本)9781479986965
stacked autoencoders enjoy their popularization with the prosperity of deep learning in recent years. However, relative studies seldom exploit the intrinsic information buried in the interrelations between the samples with respect to deep networks. Regarding this, the manifold regularization is introduced to analyze the neighborhood of each training sample, which leads to a manifold regularized stacked autoencoder hierarchical framework with deep multilayer substructures. A series of experiments are conducted upon MNIST and YaleFaces using locally linear embedding as the manifold regularization module. The results show that neighborhood analysis should be combined with stacked autoencoders to achieve some notable promotions of their performances.
Fisher discriminant analysis (FDA) is an effective fault classification method for complicated industrial processes. However, it builds the classifier only based on the labeled training data and ignores vast amounts o...
详细信息
ISBN:
(纸本)9781728140940
Fisher discriminant analysis (FDA) is an effective fault classification method for complicated industrial processes. However, it builds the classifier only based on the labeled training data and ignores vast amounts of unlabeled data which can provide important data distribution information. To well utilize the unlabeled data for better fault classification performance, this paper presents a semisupervised FDA (SeFDA) method which integrates the stacked autoencoder (SAE) for intrinsic feature extraction and applies the majority voting based ensemble modeling strategy for a more precise solution. Different to the traditional FDA method, the proposed method builds the classifier by considering the labeled and unlabeled samples simultaneously. SeFDA first applies the SAE on the whole data samples to extract the data features. Then the data features corresponding to the labeled samples are used to train a fault discriminant model by FDA. Considering that the single SAE network training is easily affected by the initial weight values, a majority voting based ensemble modeling strategy is applied to combine the results from multiple classifiers. The case studies on the simulated Tennessee Eastman process and the real sucker rod pumping system show that the modified SeFDA method is superior to the basic FDA in terms of fault classification performance.
Unsupervised feature learning with deep networks has been widely studied in the recent years. Despite the progress, most existing models would be fragile to non-Gaussian noises and outliers due to the criterion of mea...
详细信息
ISBN:
(纸本)9781479928934
Unsupervised feature learning with deep networks has been widely studied in the recent years. Despite the progress, most existing models would be fragile to non-Gaussian noises and outliers due to the criterion of mean square error (MSE). In this paper, we propose a robust stacked autoencoder (R-SAE) based on maximum correntropy criterion (MCC) to deal with the data containing non-Gaussian noises and outliers. By replacing MSE with MCC, the anti-noise ability of stacked autoencoder is improved. The proposed method is evaluated using the MNIST benchmark dataset. Experimental results show that, compared with the ordinary stacked autoencoder, the R-SAE improves classification accuracy by 14% and reduces the reconstruction error by 39%, which demonstrates that R-SAE is capable of learning robust features on noisy data.
Deep learning has become a popular tool for fault detection in industrial processes to learn complex nonlinear ***,the features extracted from most traditional deep networks usually ignore the geometric characters and...
详细信息
ISBN:
(数字)9789887581536
ISBN:
(纸本)9781665482561
Deep learning has become a popular tool for fault detection in industrial processes to learn complex nonlinear ***,the features extracted from most traditional deep networks usually ignore the geometric characters and singularities of the process *** representative features cannot be extracted effectively,which may lead to the inaccurate modeling and is not beneficial for the process ***,this paper proposes a feature learning method based on multifractal analysis and stacked autoencoder(MF-SAE) for fault detection in complex multivariate ***-SAE can learn high-level multifractal features from the raw data in an unsupervised *** analysis is frrst introduced to extract the multi-scale self-similar characteristics from industrial process data,in which sigmoid function is added for preprocessing the process *** to the redundant information existing in the multi-scale feature,SAE is then utilized to learn key feature from the extracted multifractal *** learned hidden feature and residual feature are provided to construct the monitoring *** fault detection performance of MF-SAE is tested on the Tennessee Eastman(TE) process.
暂无评论