Narrow Results

This paper aims to develop an algorithm to detect heart diseases through ordinary smartphones without additional equipment for cost accessibility. Among various vital signs emitted by organs, sounds can be easily observed and carry ample information. However, these sounds are small and noisy. Detecting anomalies involves great challenges in signal processing. This study presents a novel method that overcomes noises to estimate cardiovascular health. We use time-scale techniques in time series analysis to extract disease traits and suppress excessive ambient noises. Using datasets from PhysioNet, our model achieved a nearly 100% accuracy in heart disease diagnosis. Our approach also performs well under excessive noises for diseases producing heart murmurs. With heavy noise contaminated signals, training accuracy still closed to 100%, and the testing accuracy still remained around 84%.

In order to improve the accuracy and precision of online learning-based collision detection methods, an online active ensemble learning for robot collision detection (OAELRCD) is proposed in this paper. The OAELRCD consists of two key components: (1) an ensemble learning method to combine several base classifiers in order to improve the accuracy and precision of collision detection, (2) an active learning algorithm to reduce the number of training samples in order to realize online training and learning when the environment changes. We evaluate the proposed OAELRCD on one robot arm in dynamic environments with moving workspace obstacles, showing that the proposed OAELRCD outperforms state-of-the-art online learning-based method and geometric collision checkers. Compared to the state-of-the-art online learning-based method for robot collision detection in dynamic environments, the proposed OAELRCD provides noticeable improvements in TPR, AUC, Accuracy and TNR. Compared to state-of-the-art geometric collision checkers, with the proposed OAELRCD, collision checks are faster.

Credit risk evaluation has been vital for financial institutions to identify default customers and to avoid financial loss. Machine learning and data mining techniques have been adopted to develop scoring models for enhancing the prediction performance of default customers. However, it is difficult for these machine learning models for explaining the rejection or approval decision-making process to customers and other non-technical personnel. This paper presents an evidence reasoning (ER) rule-based ensemble learning approach for credit risk evaluation considering customer heterogeneity. Firstly, customers are segmented into different groups by $k$-means clustering algorithms and a two-stage weighting method is proposed to determine the significances of attributes by their discriminating powers between groups and within groups. Then, the attribute-related evidence is obtained by Bayesian statistics to represent the relationships between the attributes and credit risks, and a two-stage weighting evidential reasoning (TER) is developed as a base learner for credit scoring. Lastly, multiple base learners TERs are aggregated for evaluating customers’ credit risks. An empirical study on three credit datasets demonstrated that the proposed approach can achieve high performance with good explainability. The predicted results of the model can be well comprehended by providing the contribution of attributes and the activated rules in evidential reasoning processes.

Prediction is an important way to analyse stroke risk management. This study explored the critical influencing factors of stroke, used the classical multilayer perception (MLP) and radial basis function (RBF) machine learning (ML) algorithms to develop the model for stroke prediction. The two models were trained with Bagging and Boosting ensemble learning algorithms. The performances of the prediction models were also compared with other classical ML algorithms. The result showed that (1) total cholesterol (TC) and other nine factors were selected as principal factors for the stroke prediction; (2) the MLP model outperformed RBF model in terms of accuracy, generalization and inter-rater reliability; (3) ensemble algorithm was superior to single algorithms for high-dimension dataset in this study. It may come to the conclusion that this study improved the stroke prediction methods and contributed much to the prevention of stroke.

A prevalent method for rapid prototyping of metallic parts is gas metal arc welding (GMAW). As the input parameters impose a highly nonlinear impact on the weld bead geometry, precise estimation of the geometry is a complex problem. Therefore, in this study, a novel combination of the most powerful machine learning algorithms is selected to overcome the complexity of the problem and also reach an acceptable degree of precision. To this end, the hybrid combination of the support vector machine (SVM) and relevance vector machine (RVM) is developed based on the random forest (RF) ensemble learning approach. The models are established based on a global database of welding geometry, and the corresponding process parameters obtained are based on a set of experiments. Performance evaluation between RVM, SVM, and the proposed model was performed based on the coefficient of determination ($R^2$) and the ratio of root means square error (RMSE) to the maximum measured outputs ($\mathrm{\text{RMSE}}$/$y_{\mathrm{\text{max}}}$). The RF-based RVM-SVM model obtained 0.9725 and 0.8850 for $R^2$ and 0.0257 and 0.0447 for $\mathrm{\text{RMSE}}$/$y_{\mathrm{\text{max}}}$ in predicting the height and width of the bead, respectively. The result clearly showed the effectiveness of the proposed model in predicting the GMAW trend.

The statistics about the open reading frames, the base compositions and the properties of the predicted secondary structures have potential to address the problem of discriminating coding and noncoding transcripts. Again, the Next Generation Sequencing platform, RNA-seq, provides us bounty of data from which expression profiles of the transcripts can be extracted which urged us adding a new set of dimension in this classification task. In this paper, we proposed CNCTDiscriminator — a coding and noncoding transcript discriminating system where we applied the integration of these four categories of features about the transcripts. The feature integration was done using both hypothesis learning and feature specific ensemble learning approaches. The CNCTDiscriminator model which was trained with composition and ORF features outperforms (precision 83.86%, recall 82.01%) other three popular methods — CPC (precision 98.31%, recall 25.95%), CPAT (precision 97.74%, recall 52.50%) and PORTRAIT (precision 84.37%, recall 73.2%) when applied to an independent benchmark dataset. However, the CNCTDiscriminator model that was trained using the ensemble approach shows comparable performance (precision 89.85%, recall 71.08%).

Molecular recognition features (MoRFs) usually act as “hub” sites in the interaction networks of intrinsically disordered proteins (IDPs). Because an increasing number of serious diseases have been found to be associated with disordered proteins, identifying MoRFs has become increasingly important. In this study, we propose an ensemble learning strategy, named MoRFPred_en, to predict MoRFs from protein sequences. This approach combines four submodels that utilize different sequence-derived features for the prediction, including a multichannel one-dimensional convolutional neural network (CNN_1D multichannel) based model, two deep two-dimensional convolutional neural network (DCNN_2D) based models, and a support vector machine (SVM) based model. When compared with other methods on the same datasets, the MoRFPred_en approach produced better results than existing state-of-the-art MoRF prediction methods, achieving an AUC of 0.762 on the VALIDATION419 dataset, 0.795 on the TEST45 dataset, and 0.776 on the TEST49 dataset. Availability: http://vivace.bi.a.u-tokyo.ac.jp:8008/fang/MoRFPred_en.php.

RNA-binding proteins (RBPs) have crucial roles in various cellular processes such as alternative splicing and gene regulation. Therefore, the analysis and identification of RBPs is an essential issue. However, although many computational methods have been developed for predicting RBPs, a few studies simultaneously consider local and global information from the perspective of the RNA sequence. Facing this challenge, we present a novel method called DeepBtoD, which predicts RBPs directly from RNA sequences. First, a $k$-BtoD encoding is designed, which takes into account the composition of $k$-nucleotides and their relative positions and forms a local module. Second, we designed a multi-scale convolutional module embedded with a self-attentive mechanism, the ms-focusCNN, which is used to further learn more effective, diverse, and discriminative high-level features. Finally, global information is considered to supplement local modules with ensemble learning to predict whether the target RNA binds to RBPs. Our preliminary 24 independent test datasets show that our proposed method can classify RBPs with the area under the curve of 0.933. Remarkably, DeepBtoD shows competitive results across seven state-of-the-art methods, suggesting that RBPs can be highly recognized by integrating local $k$-BtoD and global information only from RNA sequences. Hence, our integrative method may be useful to improve the power of RBPs prediction, which might be particularly useful for modeling protein-nucleic acid interactions in systems biology studies. Our DeepBtoD server can be accessed at http://175.27.228.227/DeepBtoD/.

Antimicrobial peptides (AMPs), as the preferred alternatives to antibiotics, have wide application with good prospects. Identifying AMPs through wet lab experiments remains expensive, time-consuming and challenging. Many machine learning methods have been proposed to predict AMPs and achieved good results. In this work, we combine two kinds of word embedding features with the statistical features of peptide sequences to develop an ensemble classifier, named EnAMP, in which, two deep neural networks are trained based on Word2vec and Glove word embedding features of peptide sequences, respectively, meanwhile, we utilize statistical features of peptide sequences to train random forest and support vector machine classifiers. The average of four classifiers is the final prediction result. Compared with other state-of-the-art algorithms on six datasets, EnAMP outperforms most existing models with similar computational costs, even when compared with high computational cost algorithms based on Bidirectional Encoder Representation from Transformers (BERT), the performance of our model is comparable. EnAMP source code and the data are available at https://github.com/ruisue/EnAMP.

Quantum machine learning has shown advantages in many ways compared to classical machine learning. In machine learning, a difficult problem is how to learn a model with high robustness and strong generalization ability from a limited feature space. Combining multiple models as base learners, ensemble learning (EL) can effectively improve the accuracy, generalization ability and robustness of the final model. The key to EL lies in two aspects, the performance of base learners and the choice of the combination strategy. Recently, quantum EL (QEL) has been studied. However, existing combination strategies in QEL are inadequate in considering the accuracy and variance among base learners. This paper presents a hybrid EL framework that combines quantum and classical advantages. More importantly, we propose an efficient combination strategy for improving the accuracy of classification in the framework. We verify the feasibility and efficiency of our framework and strategy by using the MNIST dataset. Simulation results show that the hybrid EL framework with our combination strategy not only has a higher accuracy and lower variance than the single model without the ensemble, but also has a better accuracy than the majority voting and the weighted voting strategies in most cases.

Binary classification with an imbalanced dataset is challenging. Models tend to consider all samples as belonging to the majority class. Although existing solutions such as sampling methods, cost-sensitive methods, and ensemble learning methods improve the poor accuracy of the minority class, these methods are limited by overfitting or cost parameters that are difficult to decide. This paper proposes a hybrid approach with dimension reduction that consists of data block construction, dimensionality reduction, and ensemble learning with deep neural network classifiers. The performance is evaluated on eight imbalanced public datasets in terms of recall, G-mean, AUC, F-measure, and balanced accuracy. The results show that the proposed model outperforms state-of-the-art methods.

Nutrition is an essential component in agriculture worldwide to assure high and consistent crop yields. The leaves frequently present signs of nutritional deficiencies in rice crops. A nutritional deficiency in the rice plant can also be diagnosed based on the leaf color and form. Image categorization is an effective and rapid method for analyzing such conditions. However, despite significant success in image classification, Ensemble Learning (EL) has remained elusive in paddy nutrition analysis. Ensemble learning is a technique for deliberately constructing and combining numerous classifier models to tackle a specific computational issue. In this work, we investigate the preciseness of several uncertain deep learning algorithms to detect nutritional deficits in rice leaves. Through soil and agricultural studies, around 2000 images of rice plant leaves were collected encompassing complete nutritional and about five divisions of nutrient deficiencies. The image proportion for learning via training, validation via evaluation, and testing phase were split into 4: 2: 2. For this, an EL method is chosen for the diagnosis and classification of nutritional deficits. Here, EL procedures are considered as a hybrid classification model that integrates CapsNET (Capsule network) and GCN (Graph Convolutional Neural) networks to evaluate the classification. The hybrid classification effectiveness was verified through color and lesion features which were compared with standard machine learning techniques. This research shows that EL strategies can effectively detect nutritional deficits in paddy. Furthermore, the suggested hybrid classification model achieved a better accuracy rate, along with sensitivity and specificity rates of 97.13%, 97.22%, and 96.47% correspondingly.

Increasing availability of the multi-scale market data exposes limitations of the existing quantitative models such as low accuracy of the simplified analytical and statistical frameworks as well as insufficient interpretability and stability of the best machine learning algorithms. Boosting was recently proposed as a simple and robust framework for intelligent combination of the clarity and stability of the analytical and parsimonious statistical models with the accuracy of the adaptive data-driven models. Encouraging results of the boosting application to symbolic volatility forecasting have also been reported. However, accurate forecasting does not always warrant optimal decision making that leads to acceptable performance of the portfolio strategy. In this work, a boosting-based framework for a direct trading strategy and portfolio optimization is introduced. Due to inherent adaptive control of the parameter space dimensionality, this technique can work with very large pools of base strategies and financial instruments that are usually prohibitive for other portfolio optimization frameworks. Unlike existing approaches, this framework can be effectively used for the coupled optimization of the portfolio capital/asset allocation and dynamic trading strategies. Generated portfolios of trading strategies not only exhibit stable and robust performance but also remain interpretable. Encouraging preliminary results based on real market data are presented and discussed.

With the explosion of multimedia data, semantic event detection from videos has become a demanding and challenging topic. In addition, when the data has a skewed data distribution, interesting event detection also needs to address the data imbalance problem. The recent proliferation of deep learning has made it an essential part of many Artificial Intelligence (AI) systems. Till now, various deep learning architectures have been proposed for numerous applications such as Natural Language Processing (NLP) and image processing. Nonetheless, it is still impracticable for a single model to work well for different applications. Hence, in this paper, a new ensemble deep learning framework is proposed which can be utilized in various scenarios and datasets. The proposed framework is able to handle the over-fitting issue as well as the information losses caused by single models. Moreover, it alleviates the imbalanced data problem in real-world multimedia data. The whole framework includes a suite of deep learning feature extractors integrated with an enhanced ensemble algorithm based on the performance metrics for the imbalanced data. The Support Vector Machine (SVM) classifier is utilized as the last layer of each deep learning component and also as the weak learners in the ensemble module. The framework is evaluated on two large-scale and imbalanced video datasets (namely, disaster and TRECVID). The extensive experimental results illustrate the advantage and effectiveness of the proposed framework. It also demonstrates that the proposed framework outperforms several well-known deep learning methods, as well as the conventional features integrated with different classifiers.

Online anomaly detection for stream data has been explored recently, where the detector is supposed to be able to perform an accurate and timely judgment for the upcoming observation. However, due to the inherent complex characteristics of stream data, such as quick generation, tremendous volume and dynamic evolution distribution, how to develop an effective online anomaly detection method is a challenge. The main objective of this paper is to propose an adaptive online anomaly detection method for stream data. This is achieved by combining isolation principle with online ensemble learning, which is then optimized by statistic histogram. Three main algorithms are developed, i.e., online detector building algorithm, anomaly detecting algorithm and adaptive detector updating algorithm. To evaluate our proposed method, four massive datasets from the UCI machine learning repository recorded from real events were adopted. Extensive simulations based on these datasets show that our method is effective and robust against different scenarios.

Relation extraction (RE) is a crucial step for knowledge graph construction, which aims to extract meaningful relations between entity pairs in plain texts. Very few works have been studied on Chinese relation extraction (CRE) in the military field. Moreover, recent deep neural network-based methods have achieved considerable performance but still suffer from three inherent limitations, including overlapping of entities, imbalanced data and the ambiguity. Therefore, this work investigates a novel Multi-Grained Lattice Transformer (MGLT), which leverages external information of lexicon and word sense tailored for CRE. In MGLT, self-matched lexicon words and related word senses are fused through a cross-transformer mechanism to alleviate the ambiguity in texts. The finally enriched sequence representation in MGLT captures the relatedness between the head entity and the tail one, which is helpful to alleviate the overlapping of entities. Experimental results on two benchmark datasets and a self-developed dataset constructed from online military news show that the proposed MGLT achieves state-of-the-art (SOTA) performance. Compared with other typical baselines, MGLT achieves better area under curve (AUC) and $F1$-score by up to 10.46% and 6.90%, respectively. We further demonstrate the effectiveness of using ensemble learning to fully exploit complementary information from multiple MGLT-based base learners to improve the overall performance for imbalanced data classification on the military dataset. Such results indicate that the proposed ensemble learning model is effective and robust to be applied in practical applications.

Android is the most well-known portable working framework having billions of dynamic clients worldwide that pulled in promoters, programmers, and cybercriminals to create malware for different purposes. As of late, wide-running inquiries have been led on malware examination and identification for Android gadgets while Android has likewise actualized different security controls to manage the malware issues, including a User ID (UID) for every application, framework authorizations. In this paper, we advance and assess various kinds of machine learning (ML) by applying ensemble-based learning systems for identifying Android malware related to a substring-based feature selection (SBFS) strategy for the classifiers. In the investigation, we have broadened our previous work where it has been seen that the ensemble-based learning techniques acquire preferred outcome over the recently revealed outcome by directing the DREBIN dataset, and in this manner they give a solid premise to building compelling instruments for Android malware detection.

In this study, an ensemble neural network is proposed based on different feature subsets method in order to forecast the world crude oil spot price. To this end, a number of experts in database gathering and appropriate time delays were interviewed to forecast 1-step ahead of the crude oil spot price. Subsequently, different features subsets were generated randomly, each of which was then used for each of the basic classifiers. Then, three-layered feed-forward neural network models were used to model each of the basic classifiers. Finally, the prediction results of all basic classifiers were combined with a single layer perceptron neural network to formulate an ensemble output for the original crude oil price series. In order to verify and evaluate the presented method, one of the main crude oil price series, i.e. WTI crude oil spot price, was used to test the effectiveness of the proposed method. Empirical results provided evidence for the effectiveness of the proposed ensemble learning method compared to linear and nonlinear models.

The stock market is a complex, evolving, and nonlinear dynamic system. Forecasting stock prices has been regarded as one of the most challenging applications of modern time series forecasting. This paper proposes a novel multi-input Hamacher-ANFIS (adaptive network-based fuzzy inference system based on Hamacher operator) ensemble model to forecast stock prices in China’s stock market and achieve good prediction performance. We selected five stocks with the largest total market capitalization from the Shanghai and Shenzhen Stock Exchanges, measured their historical volatility over the same time period, and weighed the performance of each stock forecasting model based on the above volatility. Then, the experiment was repeated 100 times for each data set, and we calculated the comprehensive $R^2$ of the testing set according to the weight that we obtained earlier. The statistical test of the experimental results shows that: (1) In terms of comprehensive $R^2$ of the stock price, the multi-input Hamacher-ANFIS model is superior to other conventional models; (2) when compared with the nonensemble forecasting strategy, the ensemble strategy of the Hamacher-ANFIS model has significant advantages.

This work addresses the issues of noise and tissue appearance fluctuations in histopathology image classification by using a novel deep ensemble method. The experiment’s images were inherently noisy; however, the proposed approach includes features that allow for noise to be effectively encountered while classification tasks are being completed. This integration streamlines the categorization process by eliminating the requirement for a separate denoising phase. This approach encompasses studies on two types of noise, namely Gaussian and Rician, both commonly encountered in histopathological images. Remarkably, our proposed model demonstrated effectiveness in handling both types of noise, yielding satisfactory performance across diverse noise conditions. The proposed ensemble model achieves an accuracy of 83.74%, an $F$1-score of 81.72%, an $F$2-score of 81.04%, and an MCC of 83.99% for the highest level of rician noise. The proposed approach improves classification resilience and accuracy by combining the output of several deep-learning models. It does this by increasing the $F$2-score for malignant classes by 3–5%, which helps to reduce False Negatives. This approach differs from current technology and has promising implications for the diagnosis and treatment of breast cancer. Compared to other approaches, our suggested model performs better at higher noise levels. LIME and saliency map integration improve the interpretability of model decisions, which in turn improves classification accuracy and decision clarity. These features emphasize the adaptability and resilience of the suggested method, highlighting it as a potential instrument for enhancing the results of breast cancer diagnosis and therapy in clinical settings. The workload for pathologists is lessened, and diagnostic consistency and accuracy are improved through automation of the classification process.