Multivariate time series data are essential for analyzing financial market dynamics, particularly in the cryptocurrency sector. Predicting Bitcoin price and volatility patterns is challenging yet vital for robust investment strategies. By combining empirical and variational mode decomposition (EMD/VMD) with evolutionary-optimized temporal windows, advanced multivariate models extract temporal patterns and improve short-term forecasting performance in cryptocurrency markets. This study develops a multivariate machine learning model to forecast Bitcoin price and volatility patterns. Prices are forecast using single- and multistep predictions with minimized prediction error. The proposed five-stage framework first splits Bitcoin raw data into training, validation, and test sets, then causally extracts features via EMD/VMD. Data are then windowed. The model is trained using support vector regression (SVR), long short-term memory (LSTM), and random forest (RF), with input patterns optimized through an evolutionary algorithm to derive a selective zero–one matrix. To predict Bitcoin volatility patterns, the model is reformulated as a two-class classification task employing the same framework with optimized features and evaluated on test data. Using real-world multivariate Bitcoin time-series data (2019–2024), the proposed model delivers outstanding short-term price forecasting, with the RF method outperforming SVR and LSTM. It accurately predicts prices with low test errors. Additionally, the classification approach can identify robust and meaningful volatility trends. Future enhancements may extend its applicability to long-term predictions and other cryptocurrencies, advancing data-driven financial strategies.

In this research, an extended random forest algorithm for multivariate time series several steps forecasting is proposed. Peoposed method consists input layer and hidden layers involves random forest. In addition, a new algorithm is proposed in the third step to an ensemble of the tree’s outputs with the concept of correlation with the final results to reduce redundancy. In the output layer, a new algorithm is proposed to learn the weight of each random forest tree to calculate the result. Beijing PM25 and Italian air quality, were used to evaluate the proposed method. The results of the proposed method in this research were compared with the other state-of-the-art methods like deep learning and deep forest. We evaluated our proposed model based on evaluation metrics RMSE, MAE and MAPE and achieved good results. According to the results, the proposed method on the Beijing PM2.5 dataset’s RMSE and MAE value respectively are 40.97 and 24.81 for the average forecast for the next 1–6 h, 2.51 and 0.48 less than the best of the others. The average forecast for the next 1–3 h are 2.71 and 1.42 less than the best value of the others and are equal to 31.64 and 18.09. On the Italian air quality dataset, the RMSE, MAE and MAPE value for the next 1-h forecast are 0.6109, 0.4224 and 33.80 and better than the others.

Background: Classifying ultrasound images is one of the quickest approaches to diagnose breast cancer. In recent years, the medical community has increasingly turned to deep learning-based methods, which have demonstrated superior performance compared to other existing methods for most computer vision tasks. Since deep learning approaches suffer from over-fitting problem, the main challenge in the medical image analysis is the lack of enough available dataset to train deep learning models to extract efficient features. Many existing methods rely on data augmentation techniques to address the issue of over-fitting especially in imbalanced datasets. However, the process of finding appropriate data augmentation methods often involves trial and error, and it can be time-consuming. Therefore, there is a need to develop an algorithm that can automatically generate effective features, and minimize the reliance on a large amount of data to overcome over-fitting.Method: In this study, we introduced a novel ensemble meta-feature space generator (EMFSG-Net) for the classification of breast ultrasound images. We first employed a transfer learning approach to obtain initial features from raw images. Then we applied our method to create a more efficient feature space from the initial features. Our method is inspired by ensemble methods that aim to minimize bias and variance errors. Ensemble methods are widely used in previous works based on classification models. The novelty of our approach is to adapt these techniques to regression-based models, which enables us to construct a meta-feature generator capable of producing an effective feature space. This method offers an effective solution to address the overfitting problem commonly encountered in deep learning for medical image processing. By obtaining more suitable features, it significantly improves the classification performance. Notably, this method eliminates the need for data augmentation, which saves time by avoiding the search for appropriate data augmentation techniques. This approach demonstrates its efficacy specifically in addressing imbalanced datasets, which are characterized by a disproportionate distribution of class labels.Result: We have evaluated our model on BUSI dataset, a collection of breast ultrasound images that is widely used in scientific researches to detect breast cancer. We have achieved the accuracy of 97.96% and F1-score of 96.2% on the dataset.Conclusion: The experimental results show that the proposed model has increased 4.8% of the classification F1-score compared to other feature extraction and deep learning methods.

Evidence suggests that human emotions can be detected using Electroencephalography (EEG) brain signals. Recorded EEG signals, due to their large size, may not initially perform well in classification. For this reason, various feature selection methods are used to improve the performance of classification. The nature of EEG signals is complex and unstable. This article uses the Empirical Mode Decomposition (EMD) method, which is one of the most successful methods in analyzing these signals in recent years. In the proposed model, first, the EEG signals are decomposed using EMD into the number of Intrinsic Mode Functions (IMF), and then, the statistical properties of the IMFs are extracted. To improve the performance of the proposed model, using the RBF kernel and Least Absolute Shrinkage and Selection Operator (LASSO) feature selection, an effective subset of the features that have changed the space is selected. The data are then clustered, and finally, each cluster is classified with a decision tree and random forest and KNN. The purpose of clustering is to increase the accuracy of the classification, which is achieved by focusing each cluster on a limited number of classes. This experiment was performed on the DEAP dataset. The results show that the proposed model with 99.17% accuracy could perform better than recent research such as deep learning and show good performance. In the latest years, with the development of the BCI system, the demand for recognizing emotions based on EEG has increased. We provide a method for classifying clustered data that is efficient for high accuracy.

Background: Brain-computer interface (BCI) technology has many applications in various scientific fields, such as used in communication (speech recognition). The data of imagery speech has been collected in electroenceph-alogram (EEG) signals. Method: In this paper, we propose an approach for EEG feature extraction of imagined speech with high accuracy and efficiency. In this way, we improve the common spatial pattern (CSP) binary algorithm to multiclass level in two parts 'One-vs-One' and 'One-vs-All'. The "Kara One" dataset is used in this research that includes EEG signals of thirteen subjects with twelve trials and sixty-four channels for any four English words signals and seven English phonemes signals. Results: We compared our proposed CSP to other imagined speech feature methods. The classification accuracy of the second part of the proposed method is 97.34% in the subject-wise overall model which is 19.97% better than the best previous result. Conclusion: We have obtained the highest classification accuracy for sixty-four channels, which is the highest accuracy ever achieved using this database. Our proposed model is ready to be tested with more EEG data. This proposed work, which includes an ensemble method for classifying speech imagery words, can greatly contribute to intuitive BCI development using silent speech.

Background: Brain-Computer Interface (BCI) based on Motor Imagery (MI) is one of the emerging technology that has been used in smart healthcare applications that help disables connect with the real world by imagining a specific movement in the brain.Method: In this study, some novel features are presented for the classification of electroencephalography (EEG) signals which are named CSP-Ph-PS. This feature is extracted based on the phase space reconstructed by the filtered signal using CSP. In the proposed method, we fit Poincare acute accent sections in phase space to analyze data tra-jectory. The hyperparameters of Phase Space Reconstruction (PSR) and Poincare acute accent sections are learned with Evolutionary Algorithm (EA). Finally, statistical features extracted from the Poincare acute accent intersection points are given for classification.Result: The proposed method is implemented on two public datasets that are BCI Competition III Dataset Iva and BCI Competition IV Dataset 1 and have been achieved an accuracy of 89.76% and 71.87% on these two datasets, respectively. These are 3.16% and 1.25% better than the best previous results, respectively.Conclusion: The experimental results show that the proposed method can separate different classes of signals and has increased the classification accuracy compared to other feature extraction and deep learning methods.

The electroencephalogram (EEG) is one of the most common methods for studying epileptic seizures. Seizure detection has been a well-known subject in neuroscience. According to previous research projects, it is more effective to extract features from the EEG to detect seizures rather than analyzing the raw EEG signals. Therefore, a new feature extraction method is suggested in this work. In this paper, new features are proposed to classify different brain states in EEG records of patients with epilepsy disorder. In previous works, some of the proposed features were based on the ellipses' data density in phase space which is the motivation of this paper. The innovation here is to use clusters with no specific shapes instead of ellipses to extract improved features. A clustering method is performed on EEG signals in phase space. The densities of data in each cluster are considered as features, and these features are given to a classifier as inputs to classify different brain signals. The investigation is performed on the CHB-MIT scalp EEG dataset. In the binary classification scenario, the achieved results include a sensitivity of 94.94%, a specificity of 94.94%, and a FPR of 0.051 per hour. On the other hand, for the multiclass classification, the specificity is 92.33% and the FPR is 0.077 per hour. These performance metrics highlight the effectiveness of the proposed feature extraction method in generating robust features for classifying brain states. The achieved results demonstrate its superiority over the methods used in previous studies. By extracting meaningful and discriminative features from the EEG signals, the classification model can accurately differentiate between different brain states. This suggests that the proposed method holds promise for improving the accuracy and reliability of brain state classification. The proposed technique can assist neuroscientists in easily distinguishing different states of the brain before the seizure onset.

Electroencephalogram (EEG) signals show the electrical activity of the brain, which are one of the inputs of the brain-computer interface (BCI). The BCI provides the communication path between the brain and the computer. One of the critical applications of BCI is Motor imagery (MI). MI is a mental process that a person practices or simulates a particular movement without physically acting. BCI allows the person to communicate with their environment independently of peripheral muscles and nerves, using EEG brain signals by assistive devices such as wheelchairs, robotic arms, and computers. In this paper, a space learning concept is proposed for EEG motor imagery signal classification. Our innovation in the proposed method is to increase and then, reduce the data dimensions, which has led to learning the efficient space for signals classification. It is based on two techniques: Multi-Kernel Learning (MKL) and dimension reduction. The composite kernel is made of a combination of four kernels by The Heuristic MKL Algorithm. This algorithm uses heuristic rules to estimate the weight of kernels with high accuracy and very little computational complexity. The weight associated with each base kernel and its parameters is calculated by the Equilibrium Optimizer. Dimensions of data are reduced to avoid the curse of dimensions. In this step, the number of dimensions of reduced space and the mapping matrix are learned to reduce the dimensions of data linearly. We selected ELM, KNN, and SVM classifiers for classification. The BCI Competition dataset was used for evaluation, which consists of five subsets aa, al, ay, aw, av, and two classes of the right hand and right foot. The proposed method with the ELM was improved the average classification accuracy and standard deviation by 3.9% and 2.28, respectively, and achieved 91.4% accuracy. The lower standard deviation than other methods shows that our method is more robust than all other methods to subject variety. The proposed method is compared with twelve state-of-the-art methods and has shown higher accuracy than other methods such as the deep convolutional neural networks. The results show the superiority of the proposed method over other methods in the Wilcoxon signed test.

This paper presents an approach for identifying fingerprints through the extraction of geometric and statistical features of characteristic Minutiae. The proposed approach is in accordance with statistical features to extract important points from the skeleton of a fingerprint's image. Through the addition of geometric features as a kind of preprocessing to this approach, the images are divided into distinct regions. In this approach, statistical parameters like min, max, mean and standard deviation are applied in order to compute the general abstract of the features. Another achievement in this article is the presentation of a similarity measure in identification tools which is only used for methods based on matching patterns. The Optimized version of the proposed method achieved near zero EER percentage in some of the datasets.

One of the most challenging issues in pattern recognition is the data attribution selection process. Feature selection plays a key role in solving problems with high-dimensional data and is a fundamental step in pre-processing many classifications and machine learning problems. The feature selection method reduces the amount of data and increases the category precision. Unrelated data, which can lead to inappropriate classification, is thus removed to obtain fewer features. In this paper, the Binary Gray Wolf Optimization algorithm uses the Wrapper method for feature selection. The transfer function is an essential part of BGWO to map a continuous value to a binary value. In this study, eight transfer functions are divided into two families: S-shaped and V-shaped. Previous research has used only one transfer function for the whole algorithm, and all wolves in the whole algorithm deal with this transfer function. In this paper, each wolf has its own transfer function. Because algorithms are evolutionary meta-innovations and can optimize themselves, each wolf can play a role in the whole algorithm at any stage while optimizing itself and adapting to its community, and not just depend on one transfer function. In the proposed method, eight transfer functions are divided into two families, S-shaped and V-shaped. This article proposes two approaches for learning the transfer function, by selecting the transfer function and the slope of these functions. In the first approach, we add three or two binary bits to the initial population. If two bits are added, four modes of the transfer function are available, and if three bits are added, eight transfer functions are achievable. These bits are used as a criterion for selecting a predefined transfer function for each wolf. So, in the proposed method, each wolf has its transfer function. During the implementation of the algorithm, the wolves update their position according to the evaluation function and learning. In the second approach, ten or twenty-one binary bits are added to the initial population. If ten binary bits are used, we will have a transfer function, and 2^10 coefficient modes are available for the slope of the transfer function. If twenty-one binary bits are used, we have two transfer functions available. So, there are 2^10 modes for the gradient of the transfer function. These bits are used as a criterion for selecting the transfer function and the coefficient affecting the slope of these functions. In both ideas, after each iteration of the algorithm, the position of the wolves is updated and based on the evaluation function, the alpha wolf is identified and the transfer function is selected. With subsequent iterations, the algorithm learns and optimizes the transfer function to achieve the best feature selection with the smallest error. Experimental results on ten UCI datasets show that selecting the obtained feature subset with high classification accuracy is efficient.

In this study, a new method based on chaos functions, and the evolutionary algorithm is proposed for image encryption. Chaos functions are used in this method because of the random occurrence and the sensitivity to the initial values to make the encryption method as secure as possible. Also to enhance the entropy of the image, an evolutionary algorithm is used to select the best layout and mapping. For this purpose, the image is decomposed first. The image components are then disrupted using the evolutionary algorithm, coding rules, and logistic mapping whose initial value is obtained from a hash function. The results show that the proposed method has good speed due to the use of simple operators such as Addition and XOR. Also, since a 256-bit hash function is used in this case and a high search space is generated for the evolutionary algorithm, the algorithm shows good resistance to the types of attacks. Moreover, due to the uncertainty of the decryption algorithm and the generation of a single-use code for each execution of the algorithm, the proposed encryption algorithm offers high security and resistance against differential attacks and plaintext attacks.

One of the most important issues in engineering is to find the optimal global points of the functions used. It is not easy to find such a point in some functions due to the reasons such as large number of dimensions or inability to derive them from the function. Also in engineering modeling, we do not have the relationships of many functions, but we can input and output them as a black box. Therefore, the meta-heuristic algorithms are presented. In this paper, a meta-heuristic algorithm based on the behavior of vortices in fluid physics is presented. Technically, the algorithm is made up of vortices. Each vortex contains some particles. The particles move by the presented rotation matrix. This movement causes the local search. Also by selecting another vortex through the selection algorithm, each vortex attempts to escape the local optima and reach the global optima. The algorithm will explore and exploit the given function using its operators. Another innovation of this paper is the introduction of two new evaluation criteria for optimization algorithms. These two criteria show the behavior and convergence of algorithms along the way to reach the global optimal point or fall into the local optima. The proposed algorithm has been implemented, evaluated and compared with the numerical optimization state of the art algorithms. It was observed that the proposed method was able to achieve better results than most of the other methods in the major of twenty-four standard functions in different dimensions. (All codes available at http://web.nit.ac.ir/ h.omranpour/.).

Epilepsy is one of the world’s most common neurological disorders. Timely diagnosis of this disease improves the quality of life of patients. In this research, we used deep learning to diagnose and predict epileptic seizures. While Long Short Term Memory (LSTM) learn the concept of time and Convolutional Neural Network (CNN) learn images well, Convolutional Long Short Term Memories (ConvLSTMs) as a new type of LSTMs use both capabilities. How to prepare the input is important and effective in using deep neural networks. The use of raw signals also forces us to make full use of time-domain features. We employed the short-time Fourier transform (STFT) to use both time and frequency domain information. But the output images have a fixed resolution due to the fixed size of the window. We solved this problem by calculating STFT with different window sizes and adding a third dimension. Then, with rotation, we put the dimensions in positions appropriate to their meaning. So we provided a set of images that convey the concept of time to ConvLSTMs to learn the signal pattern and generalize it. We tested our deep learning model on dataset from the University of Bonn in Germany. We compared the findings with those obtained from the other state-of-the-art models. The obtained accuracies demonstrate that the proposed model is both effective and reliable.

Ensemble Kernel Learning Model for Prediction of Time Series Based on the Support Vector Regression and Meta Heuristic Search

Nowadays time series analysis is an important challenge in engineering problems. In this paper, we proposed the Comprehensive Learning Polynomial Autoregressive Model (CLPAR) predict linear and nonlinear time series. The presented model is based on the autoregressive (AR) model but developed in a polynomial aspect to make it more robust and accurate. This model predicts future values by learning the weights of the weighted sum of the polynomial combination of previous data. The learning process for the hyperparameters and properties of the model in the training phase is performed by the metaheuristic optimization method. Using this model, we can predict nonlinear time series as well as linear time series. The intended method was implemented on eight standard stationary and non-stationary large-scale real-world datasets. This method outperforms the state-of-the-art methods that use deep learning in seven time series and has better results compared to all other methods in six datasets. Experimental results show the advantage of the model accuracy over other compared methods on the various prediction tasks based on root mean square error (RMSE).

Background and Objectives: Time series classification (TSC) means classifying the data over time and based on their behavior. TSC is one of the main machine learning tasks related to time series. Because the classification accuracy is of particular importance, we have decided to increase it in this research.Methods: In this paper, we proposed a simple method for TSC problems to achieve higher classification accuracy than other existing methods. Fast Fourier transform is a method that uses in raw time series data preprocess. In this study, we apply the fast Fourier transform (FFT) over the raw datasets. Then we use the polar form of a complex number to create a histogram. The proposed method consists of three steps: preprocessing using FFT, feature extraction by histogram computation, and decision making using a random forest classifier.Results: The presented method was tested on 12 datasets of the UCR time series classification archive from different domains. Evaluation of our method was performed using k-fold cross-validation and classification accuracy. The experimental results state that our model has been achieved classification accuracy higher or comparable than related methods. Computational complexity has also been significantly reduced.Conclusion: In the latest years, the TSC problems have been increased. In this work, we proposed a simple method with extracted features from fast Fourier transforms that is efficient to gain more high accuracy.

One of the challenges in recognizing handwritten texts is the individual style of writing. There is the structural similarity of the different digits to each other in writing. Along with the mentioned challenge, these similarities in the text may increase and make it difficult to correctly recognize digits and numbers. In this paper, a new model for recognizing handwritten digits is presented. The proposed model is a stacking ensemble classifier. This classifier is based on the convolutional neural network (CNN) and the bidirectional long-short term memory (BLSTM). Another innovation of the model is the use of the probability vector of images class as the input of the meta-classifier layer. One of the strengths of BLSTM is the ability to learn arrays and vectors; therefore, from a technical point of view, considering the output probability vector of the first model as the input of the meta-classifier (BLSTM) improves the accuracy of the deep learning model. The reason for using stacking ensemble classification is the sameness of the main body of some Persian/Arabic digits (e.g., "2, 3, and 4"). Also, the style the author writes makes classes that are not similar in a structure similar to each other, which causes errors incorrect recognition. This model helps to recognize the correct set of input digits by examining the structure of similarities. To achieve a reliable result in the face of this challenge, this model has been tested on a large Persian/Arabic dataset to cover a wide range of writing styles from different individuals. The dataset has a total of 102,352 data which 60,000 of them are for training data and 20,000 of them are for test data in ten classes of digits that are used in this paper. The result of using this database is to improve the recognition performance of these challenging digits. In examining the dataset presented by the model, the accuracy rate of the training set was 99.98%. And the sample accuracy rate in the test set was 99.39%. That is, compared to experimenting with the convolutional neural network and other researches, the rate has increased. (All codes available at http://web.nit.ac.ir/-h.omranpour/.) (C) 2021 Elsevier B.V. All rights reserved.

In this study, an efficient method for extracting and selecting features of unrefined Electroencephalogram (EEG) signals according to the one-dimensional local binary pattern (1D-LBP) is presented. Considering that taking a correct decision on various issues particularly in the field of diagnosing diseases, such as epilepsy, is of paramount importance, a functional approach is designed to extract the optimal features of EEG signals. The proposed method is comprised of two main steps: First, extraction and selection of features is performed based on a novel improved 1D-LBP model followed by data normalization through principal component analysis (PCA); as combining 1D-LBP neighboring models and PCA (1D-LBPc2p) method. The second step includes classification using two of the best ensemble classification algorithms, that is, random forest and rotation forest. A comparative evaluation is performed between the proposed methods and 13 distinct reported approaches including uniform and non-uniform 1D-LBP. The results are demonstrating that the combining method presented in our approaches has superiority along with efficiency by providing higher accuracy compared to the other models and classifiers. The proposed method in this paper can be considered as a new method for feature extraction and selection of other kinds of EEG signals and data sets.

The electroencephalography (EEG), which is a method for monitoring the brain signals, is a common method used to diagnose the epileptic seizures. In this study, some features are presented for the classification of the brain signals. These features are based on the texture and structure of the brain signals in the phase space representation (PSR). Due to the resonance property, the data are elliptical in the phase space. Therefore, the mentioned features are based on the calculations of the data density in the ellipses. In the first method of feature extraction, the radius values of the ellipses are assumed based on the normal distribution feature. In the two other methods of the presented features, the radius values of the assumed ellipses are calculated by the optimizer. These methods of feature extraction are based on the incremental ellipses and intersecting ellipses, respectively. The density of the data in the assumed ellipses is given to the k-nearest neighbor as a feature to classify the epileptic seizure and seizure-free EEG signals. The intended method was implemented and investigated on two databases of the Bonn university of Germany and the neurology and sleep center of New Delhi. The results indicate that the proposed features are strong tools for separation and diagnosis of this type of signal and have higher accuracy compared to the other classic and updated methods. The extraction speed of the presented features was higher in the test phase compared to the other methods. (C) 2020 Elsevier B.V. All rights reserved.

In this paper, a new mobile robot mapping algorithm inspired from the functionality of hippocampus cells is presented. Place cells in hippocampus can store a map of the environment. This model fuses odometry and vision data based on dimensionality reduction technique, hierarchically. These two types of data are first fused and then considered as inputs to the place cell model. Place cells do the clustering of places. The proposed Place cell model has two types of inputs: Grid cells input and input from the lateral entorhinal cortex (LEC). The LEC is modelled based on the dimension reduction technique. Therefore, the data that causes locations different to be inserted into the place cell from this layer. Another contribution is proposing a new unsupervised dimension reduction method based on k-means. The method can find perpendicular independent dimensions. Also, the distance of cluster centres found in these dimensions is maximised. The method was compared with LDA and PCA in standard functions. Although LDA is a supervised method, the result showed that the proposed unsupervised method outperformed. To evaluate the place cells model, sequences of images collected by a mobile robot was used and similar results to real place cells achieved.