Search

Development of a reliable and intelligent traffic monitoring system is highly desired to improve the transportation safety and establish future transportation plans due to the fast growth of vehicle population. Vehicle classification is one of the most critical subsystems where existing ones suffer from privacy concerns, requirements of complicated systems, and high maintenance cost. This paper reports a novel vehicle classification method by utilizing a triboelectric sensor to accurately identify vehicles. Novelty of this method originates from using triboelectric sensor and machine learning method with important advantages over current alternatives by providing an easy installation, simple operation, noninvasive measurement, cost-effective manufacturing, and highly accurate classification. To make a classification, vehicle toys' signals were acquired from triboelectric sensor and then applied to a deep learning algorithm. The 1932 sensor output data were grouped into a set of seven vehicle toys with different wheelbases, and number of tires passing on are used to train and optimize 1D-CNN model. The utilized 1D-CNN model achieved accuracy, f1-score, precision, and recall as 96.38%, 0.9638, 0.9658, and 0.9637, respectively. [ABSTRACT FROM AUTHOR]

Deep learning methods have shown significant potential in computer vision applications within the field of civil engineering. One important area of research is the detection and classification of construction machines using convolutional neural networks (CNNs). However, a major challenge in adopting machine learning models for vision systems lies in hyperparameter optimization. Therefore, the objective of this paper is to propose a meticulous statistical approach to hyperparameter tuning for deep learning in the visual classification of construction machines. It is noteworthy that the method for selecting hyperparameters utilizes statistical concepts, including analysis of variance and the Tukey test. Moreover, three research questions were formulated to address statistical differences in performance when varying the CNN architecture, optimizer and configuration (architecture and optimizer). Two additional research questions were posed to determine the recommended hyperparameters for the task and assess performance differences for non-selected hyperparameters. For this purpose, 18 hyperparameter combinations were analyzed, encompassing three CNN architectures (DenseNet, Mobilenet and VGG16) and six optimizers (adadelta, adagrad, rmsprop, adam, adamax and sgd). The Alberta Construction Image Dataset, consisting of 2850 images of three construction machines (excavator, dump truck and concrete mixer truck), was used in the experiments. The results demonstrate that the hyperparameter configurations exhibit statistically significant differences in image classification. Furthermore, the recommended combination (DenseNet × adagrad) yielded the highest average accuracy results in both binary tests (90.0%) and multiclass classification (77.8%). [ABSTRACT FROM AUTHOR]

The communication between the human brain and the external devices can be established using Electroencephalograms (EEG)-based Brain–Computer Interface by converting the neural activities of the brain into electric signals. The EEG signals were isolated into an energy–frequency–time spectrum with Hilbert Huang transform that was used by the Deep Learning (DL)-based model to learn discriminative spectro-temporal patterns of the raw EEG signals of ten digits. This paper has two major contributions: first, create a novel dataset known as BrainDigiData of EEG signals of ten digits from (0–9) using a multi-channel EEG device. Second to propose a DL-based one-dimensional Convolutional neural network model BrainDigiCNN to classify the BrainDigiData of EEG signals of digits. The publicly available Mind Big Dataset (MBD) of digits was also used to evaluate the performance of the proposed model. The research done in this paper showed that the band-wise analysis of EEG signals in a complex scenario resulted in improved results as compared to the scenario used in the previously existing work for digit classification using EEG signals. The proposed BrainDigiCNN model achieved the highest average accuracy of 96.99%. The average classification accuracy of 98.27% was achieved for the MBD dataset of 14 channel device EMOTIV EPOC+ and 89.62% on the MBD dataset of 5-channel EMOTIV Insight. The statistical analysis of the proposed model on traditional Machine Learning (ML) classifiers using paired t-test resulted in a p-value less than 0.05 which shows the significant difference between the proposed model and ML classifiers. [ABSTRACT FROM AUTHOR]

Electric vehicles (EVs) are gaining popularity as an efficient and environment-friendly transportation alternative. However, as the number of EVs on the road increases, the demand for electric power to charge them also rises, escalating challenges for the power industry. This requires accurate profiling of EV charging patterns to make informed decisions in future integrated energy systems. To address this problem, various traditional algorithms have been used to forecast the electricity demand for EV charging in the future. Recently, AI techniques, such as neural networks, machine learning, and deep learning, have shown promise in leveraging extensive historical datasets to identify patterns and generate precise predictions. This research paper introduces a novel implementation of an XGBoost regression tree-based algorithm applied to three real-world datasets to estimate the future electric load of EVs. Experimental results demonstrate the superior performance of the proposed algorithm compared to eight state-of-the-art machine learning algorithms, evaluated using the root-mean-squared error and mean absolute error. Furthermore, the proposed algorithm outperforms previous studies in this field, highlighting its effectiveness in forecasting EV charging power requirements. [ABSTRACT FROM AUTHOR]

Surface Water Bodies (SWB) are a renewable water source crucial for maintaining ecosystems and the water cycle. The declining rate of SWB increases owing to the overutilization of these resources, especially for agriculture. A timely and accurate Surface Water Body Extraction (SWBE) is necessary for water resource conservation and planning. Recently, Deep Learning (DL), a subset of Machine Learning (ML) algorithm, got remarkable attention in SWBE. It learns inherent features directly from the images at the expense of time and data. But, the ML algorithms such as K-Nearest Neighbors (KNN), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), and eXtreme Gradient Boosting (XGB) use optimal hand-crafted features to produce better results with fewer data and time. In this paper, SWBE is performed through two steps: (1) Use of spectral indices and Gabor filters for obtaining Pixel Level Feature (PLF) maps from the multispectral image; (2) Prediction of water and non-water pixels based on PLF maps using the KNN, DT, RF, SVM, and XGB classifiers. The proposed framework has experimented with Resoucesat-2 imagery over major reservoirs in Tamil Nadu and India. The results show that the proposed PLF + XGB outperforms in accuracy, recall, F1-score, kappa, False Negative Rate, Mathews Correlation Coefficient, and mean Intersection over Union with the metric value of 0.995, 0.990, 0.983, 0.979, 0.009, 0.979, and 0.969 with other existing and proposed models. Also, the surface water extent of Bhavani Sagar and Sathanur reservoirs is predicted for 4 years (2016–2019) and the causes of surface water dynamics were analyzed. [ABSTRACT FROM AUTHOR]

The concurrence of state-of-the-art Industrial 5G, Cyber-Physical Systems, Smart-Systems, Industrial Internet of Things, and Additive Manufacturing paves the next-level digital remodeling. However, the transfiguration unwittingly tailpiece an operational onus on the smart-environment operators. The multiplicity and classes of IoT devices operating in the intelligent environment are myriad. The characterization of ingress network traffic and the accurate classification of devices is necessary to efficiently manage the devices and offer cutting-edge security solutions and quality of Service (QoS). The paper addresses these challenges by offering a novel intelligent framework for traffic classification leveraging behavioral attributes of IoT traffic. The paper's contributions to the research community are fourfold. Firstly, the paper proposes a novel IoT classification framework based on Stack-Ensemble for real-time high-volume IoT traffic. The experimental results indicate that the proposed novel Stack Ensemble model can extract the best out of base models and demonstrate an accuracy of 99.94%. The intelligent models are evaluated over multiple dimensions to project the isometric view of the model performance and the experimental results. To achieve that goal, all the performance metrics that most researchers most often miss have been elucidated. Secondly, the paper comprehends the flow-level statistical characteristics of IoT devices. Third, the paper offers the distributed, scalable, and portable framework architecture. The architecture is horizontally scalable, distributing the computational load. The framework offers an end-to-end industry-grade machine-learning pipeline and triumphs the vulnerabilities of existing solutions. Finally, the paper discusses the statistical insights into the intelligent model and the results of the experimentation study. The proposed work paves the opportunities for researchers, smart-environment operators, and developers to unfold the architecture and supplement the security solutions against cyber-attacks. [ABSTRACT FROM AUTHOR]

As data grow rapidly on social media by users' contributions, specially with the recent coronavirus pandemic, the need to acquire knowledge of their behaviors is in high demand. The opinions behind posts on the pandemic are the scope of the tested dataset in this study. Finding the most suitable classification algorithms for this kind of data is challenging. Within this context, models of deep learning for sentiment analysis can introduce detailed representation capabilities and enhanced performance compared to existing feature-based techniques. In this paper, we focus on enhancing the performance of sentiment classification using a customized deep learning model with an advanced word embedding technique and create a long short-term memory (LSTM) network. Furthermore, we propose an ensemble model that combines our baseline classifier with other state-of-the-art classifiers used for sentiment analysis. The contributions of this paper are twofold. (1) We establish a robust framework based on word embedding and an LSTM network that learns the contextual relations among words and understands unseen or rare words in relatively emerging situations such as the coronavirus pandemic by recognizing suffixes and prefixes from training data. (2) We capture and utilize the significant differences in state-of-the-art methods by proposing a hybrid ensemble model for sentiment analysis. We conduct several experiments using our own Twitter coronavirus hashtag dataset as well as public review datasets from Amazon and Yelp. For concluding results, a statistical study is carried out indicating that the performance of these proposed models surpasses other models in terms of classification accuracy. [ABSTRACT FROM AUTHOR]

Microfossils provide evidence and knowledge about the Earth, life, ecology, and biological changes, and one way to access this knowledge is through the classification of microfossil species. In traditional classification methods, paleontologists analyze each microfossil sample in detail manually under a microscope. This operation not only needs extensive time for processing, but is expensive and requires well-trained experts. In recent years, this problem has been tackled through the use of automatic classification methods that benefit from machine learning and imaging techniques. Since the costs of scanning electron microscopy and micro-computer tomography systems are quite high, we focused on obtaining high performance using a low-cost light microscopic imaging system. In this paper, we have proposed a highly effective and low-cost classification system using deep learning and shallow machine learning algorithms to identify Globotruncana species. The deep model defines the data using high-level layers, whereas the shallow models recognize low-level layers during the learning process. Initially, we collected more than 180 microscopic images for three different Globotruncana species and labeled them individually. Afterwards, we applied well-known machine learning methods to this dataset and analyzed the results in detail. Experimental results showed that Globotruncana species can be classified by deep and shallow machine learning algorithms with the high accuracy of up to 100% and 100%, respectively. The performance of the classification methods was also compared in terms of their accuracy, sensitivity, specificity, precision, and F-scores. [ABSTRACT FROM AUTHOR]

Recommender systems are becoming more essential than ever as the data available online is increasing manifold. The increasing data presents us with an opportunity to build complex systems that can model the user interactions more accurately and extract sophisticated features to provide recommendations with better accuracy. To construct these complex models, deep learning is emerging as one of the most powerful tools. It can process large amounts of data to learn the structure and patterns that can be exploited. It has been used in recommender systems to solve cold-start problem, better estimate the interaction functions, and extract deep feature representations, among other facets that plague the traditional recommender systems. As big data is becoming more prevalent, there is a need to use tools that can take advantage of such explosive data. An extensive study on recommender systems using deep learning has been performed in the paper. The literature review spans in-depth analysis and comparative study of the research domain. The paper exhibits a vast range of scope for efficient recommender systems in future. [ABSTRACT FROM AUTHOR]

Availability of synchronized measurements provides an avenue for real-time measurement-based estimation of load model parameters. The presence of noise and outliers in actual PMU data makes it essential to develop estimation approaches robust to noise and outliers. Robust load parameter estimation is identified as the major goal in this research. This paper presents a time-series ML framework for robust steady-state load parameter estimation. The time-series machine learning-based approach provides the ability to intelligently estimate parameters using just voltage and power time-series data without any specific user-defined features. The convolutional neural network in 1D form (1D-CNN) is proposed for implementing the proposed time-series ML-based framework for load parameter estimation. Presented case studies demonstrate that the proposed algorithm provides robust estimation performance in the presence of realistic noise in PMU measurements. This work also introduces a bad data preprocessing framework to further enhance parameter estimation's robustness with outliers in measurement data. A detailed statistical analysis is performed to demonstrate excellent generalizing capability and robust estimation performance of the proposed parameter estimation framework. [ABSTRACT FROM AUTHOR]

Prediction of human physical activities has become a necessity for some applications that come with the development of wearable and portable hardware such as smartwatches and smartphones. The task of Human Activity Recognition (HAR) is to recognize human physical activities, e.g., walking, sitting, and running, using the data collected from sensors, e.g., accelerometers and gyroscope. HAR is commonly applied on smart systems, such as smartphones, to serve the understanding of a user's behaviors and provide assistance to the user because of the rapid development of ubiquitous computing technology in recent years. Thus, predicting activities, such as standing, walking, sitting, during the day have become a popular topic in machine and deep learning. The aim of this study is to predict the user's activities based on context information gathered by sensors such as gyroscopes and accelerometers. The conducted classification algorithms extract features from training data and learn a classification model based on the features to predict activity. In this paper, various classical machine and deep learning techniques have been studied and compared for human activity recognition. A comparative analysis is performed between techniques in order to select the classifier with the best recognition performance. Experimental results show that established Deep Neural Network (DNN) model achieved an accuracy of up to 96.81% and mean absolute error of up to 0.03 on publicly available UCI-HAR dataset. This method has given the best performance between conducted classification methods in this study to predict human activity. [ABSTRACT FROM AUTHOR]

Efficient resource provisioning is one of the important research challenges in cloud computing. Cloud providers use auto-scaling techniques for resource provisioning to manage the fluctuating workload as well as SLA violations. Auto-scaling is necessary for optimal and efficient resource allocation to manage the workload that will reduce power consumption and will also guarantee for better quality of service (QoS). It is essential for those services which support rigorous quality of service (QoS) requirements such as minimum response time or maximum throughput. Cloud providers implement service elasticity through auto-scaling mechanisms. This paper uses a novel prediction technique for resource provisioning through the auto-scaling mechanism. The prediction is done using time series data analytics with the help of the deep learning technique (i.e., LSTM). The predictive resource provisioning technique will predict the traffic load over the server and then estimates the required number of computing resources. The estimated computing resources are provisioned through the queuing theory so that it optimizes the response time and satisfies the SLA contract between the end-user and provider. The optimal resource provisioning will also reduce the consumption of computing resources while attenuating resource over-provisioning minimizes energy consumption and reduces the infrastructure's operating costs. The experimental results show that the workload prediction through LSTM obtains better accuracy than other traditional models. [ABSTRACT FROM AUTHOR]

Data clustering is an unsupervised learning method as a pivotal technique for statistical data analysis. It is a challenging machine learning scheme that involves the grouping of data samples, especially in large databases. Deep neural networks are scalable to large-scale data and capable of learning data structure by modeling the nonlinearity. One of the famous latent generative models in this realm is the generative adversarial network (GAN). In the latent generative models, for clustering, we need the posterior corresponding to the intended model. Then, we need a variational approximation of that. To address this problem, we can maximize mutual information or minimize the KL-divergence. In this paper, to reach a more generalized inference in clustering, an ensemble approach is employed to approximate the posterior. To implement this ensemble with deep networks, we proposed a convex lower bound for the posteriors' variational approximation. To amend the generator behavior, we injected the geometrical structure of data as manifold regularization to the objective function to reach accurate statistical inference. The efficacy of the proposed method has been addressed in four benchmark data sets. The experimental results confirm our model's superiority in comparison with standard clustering algorithms and some recently developed deep methods. [ABSTRACT FROM AUTHOR]

The eMBB (enhanced Mobile BroadBand), URLLC (Ultra-Reliable, Low latency communication) and mMTC (massive Machine Type communication) are the drivers for 5G communication. To realize these use-cases, enhancing the throughput in available bandwidth is the fundamental requirement in the next-generation networks. If all these use-cases are satisfied without increasing the spectral efficiency, the day is not far when we start looking for even higher frequencies (probably 6G). Applying machine learning at all possible avenues in the physical layer will be a game-changer. In this paper, we propose a novel deep learning (DL) method for hybrid precoding to maximize the spectral efficiency. we consider a special case of the MIMO system with a single-output (MISO) and implement DL technique in hybrid precoding for perfect and imperfect Channel State Information (CSI). Though the blackbox method suits for massive MIMO systems with perfect CSI, we introduce a new deep learning method which directly outputs optimized beamforming even in imperfect CSI conditions. Simulation results show that the proposed DL-based beamformer improves spectrum throughput while being more robust to imperfect CSI over the traditional beamforming approaches. This work paves a way to implement machine learning in physical layer beamforming technique for 5G millimeter wave (mmWave) communications, thereby realizing cognition in wireless networks. [ABSTRACT FROM AUTHOR]

Heart diseases are currently a major cause of death in the world. This problem is severe in developing countries in Africa and Asia. A heart disease predicted at earlier stages not only helps the patients prevent it, but I can also help the medical practitioners learn the major causes of a heart attack and avoid it before its actual occurrence in patient. In this paper, we propose a method named CardioHelp which predicts the probability of the presence of cardiovascular disease in a patient by incorporating a deep learning algorithm called convolutional neural networks (CNN). The proposed method is concerned with temporal data modeling by utilizing CNN for HF prediction at its earliest stage. We prepared the heart disease dataset and compared the results with state-of-the-art methods and achieved good results. Experimental results show that the proposed method outperforms the existing methods in terms of performance evaluation metrics. The achieved accuracy of the proposed method is 97%. [ABSTRACT FROM AUTHOR]

Currently, it is easy to create content and share it via social media platforms such as Twitter, Facebook, and Sina Weibo. However, some problems can occur when the shared content includes untrustworthy or misleading information. Thus, researchers from different domains have tried to investigate the impact of rumors on the global community. Several machine learning approaches have been used to detect rumors at their early stage. However, the achieved accuracies demonstrate that the existing state-of-the-art rumor detection approaches still require improvement. In this paper, we propose a deep learning model based on a conventional neural network (CNN) to detect rumors spreading on Twitter. Several experiments were conducted to find the best hyperparameter settings to improve the model's performance. We compared our results with other relevant rumor detection approaches that used the same publicly available benchmark dataset to demonstrate our model's performance regarding the accuracy, precision, recall and f-measure. The results show that our CNN model outperforms all the existing approaches and achieves the best balance of recall and precision. [ABSTRACT FROM AUTHOR]

Emotion recognition from human speech is essential to understand the convoluted human nature. For any machine to accurately decipher the intended message in the speech, it must understand the emotion of spoken words. Emotions control the modulations in the speech, and these modulations may even change the context. Through this paper, we aim to propose a system which can efficiently detect the emotions from speech. The domain of emotion recognition from human speech is very complex due to highly overlapping regions of emotions, and it sometimes becomes very difficult to distinguish between two emotions just based on voice. Such ambiguity in the label assignment is responsible for low classification accuracy in existing systems. In the proposed system, we have worked on finding both the suitable feature set as well as the classifier. The proposed system achieved 29.74% increase in classification accuracy in comparison with the baseline human accuracy on the primary dataset, i.e. 'CREMA-D'. Further, we have validated on other standard datasets such as 'EmoDB', 'RAVDESS', and 'SAVEE'. 'EmoDB' is a German language dataset, while the other two are English language datasets, which is in line with the language-independent nature of our system. When compared to the current state of the art in this domain on these datasets, the proposed system gives better accuracies for most of the cases, and for some cases, it gives comparable accuracies to baseline models or existing published work. [ABSTRACT FROM AUTHOR]

Advances on seismic survey techniques require a large number of geophones. This leads to an exponential growth in the size of data and prohibitive demands on storage and network communication resources. Therefore, it is desirable to compress the seismic data to the minimum possible, without losing important information. In this paper, a stacked auto-encoder extreme learning machine (AE-ELM) for seismic data compression is proposed. First, a deep asymmetric auto-encoder is constructed, in which nonlinear activation functions are used in the encoder hidden layers and linear activation functions are utilized in the decoder layers. Second, the encoder hidden layers are connected in a cascade way, so that outputs of a hidden layer are considered as the inputs to the succeeding hidden layer. Third, the optimal weights of connections between the layers of the decoder are solved analytically. Lastly, the AE-ELMs are stacked to create the complete encoder/decoder. The extreme learning machine (ELM) is selected due to its analytical calculation of weights efficient training that is suitable for practical implementation. In this neural network, data compression is achieved by transforming the original data through the encoder layers where the size of outputs from the last encoder hidden layer is smaller than the original data size. The proposed method exhibits a comparable reconstruction quality on a real dataset but with a much shorter training duration than other deep neural networks methods. This neural network with more than 8000 hidden units achieved 1.28 × 10 - 3 of normalized mean-squared error for 10:1 of compression ratio with only 8.23 s of training time. [ABSTRACT FROM AUTHOR]

In this paper, the ability of Bayesian and convolutional neural networks (CNNs), as two different machine learning methods, to recognize Arabic handwritten words is analyzed. Our contribution is threefold. First, we describe the main highlights of the dynamic Bayesian network (DBN) architecture, especially when compared to standard Bayesian networks. For that, some structural features are extracted from word image and considered as input for different architectures of Bayesian networks (BNs) such as Naïve Bayes (NB), Tree Augmented Naïve Bayes (TAN), Forest Augmented Naïve Bayes (FAN) and Hidden Markov model (HMM). Features are extracted based on the word baseline which has been estimated to mainly cope with the problems of inclination and distortions. Decisions about word classification are then inferred using multiples models of BNs. Second, we model a deep learning architecture: a CNN that convolves learned features with input data and uses 2D convolutional layers that makes it well suited to 2D word image processing. Third, we compare the behavior of DBN–CNN and propose to combine them to exploit their advantages. Experiments are carried on the standard IFN–ENIT database. The obtained results show the relatively high accuracy of the DBN and CNN combination: 95.20% compared to the remaining models. [ABSTRACT FROM AUTHOR]