Missing values are a common issue in many high-dimensional datasets, but a majority of classification algorithms require complete data. Therefore, imputation methods are usually used to estimate and fill missing values with plausible values before using the classification algorithms to learn classifiers or using the learnt classifiers to classify unseen incomplete samples. However, good imputation methods are usually computationally intensive on high-dimensional datasets because these datasets not only have a large number of features, but also often suffer from a large number of missing values. Another approach is to use decision tree algorithms which do not need imputation, and can work directly with incomplete data. However, using decision trees to classify high-dimensional data often leads to large classification accuracy because of the curse of dimensionality. Ensemble techniques which build multiple classifiers instead of a single classifier have been widely used to improve accuracy for decision trees. This paper aims to investigate different ensemble methods to find effective and efficient ensembles of decision trees for classification with high-dimensional incomplete data. Experimental results show that the random subspace method is the most accurate ensemble. The random subspace method is also more accurate than other classification algorithms which needs to combine with imputation when working with incomplete data. Moreover, the random subspace method is much faster than the other algorithms because it can directly work on incomplete data, so does not have to spend time estimating missing values.

Classification with high-dimensional data is a significant challenge in machine learning because the abundance of features in high-dimensional data makes it difficult to identify meaningful patterns, which leads to overfitting and reduced classification performance. Moreover, the computational cost of processing high-dimensional data is often prohibitively expensive, requiring specialized hardware or optimized algorithms. Ensemble learning is a powerful machine learning technique that combines multiple models to improve classification accuracy. By aggregating the predictions of multiple models, ensemble learning can reduce overfitting, increase robustness, and improve performance on a wide range of real-world classification problems. Ensemble learning is effective for classification with high-dimensional data because it can combine multiple models to mitigate the effects of the curse of dimensionality, reduce overfitting, and enhance generalization performance. By using different learning algorithms or subsets of features, ensemble learning can improve the diversity of the models, leading to better overall performance on high-dimensional data. This paper proposes two hybrid ensemble machine learning approaches that integrate random subspace ensemble with bagging and boosting to enhance classification performance with high-dimensional data. Experimental results demonstrate that these methods significantly improve classification accuracy with highdimensional data.

Abstract Classification represents a fundamental task in machine learning and data mining. Many real-world datasets exhibit a large number of features but only a limited number of instances, which often poses challenges for conventional classification algorithms. Dealing with such datasets, one prevalent approach involves using feature selection to retain relevant features while eliminating irrelevant ones. Although evolutionary computation algorithms have been extensively used for such problems, they do not perform well with these high-dimensional datasets. To address these limitations, this paper introduces a random subspace evolutionary feature selection method that performs feature selection on small subsets of features to avoid overfitting and speed up the process and employs multiple feature subspaces to prevent stagnation in local optima during feature selection. The experimental results demonstrate that this novel method achieves higher accuracy and greater stability compared to other commonly used benchmark feature selection techniques.

Honey bees (Apis mellifera) play a very important role in agriculture thanks to their ability of plants' pollination. However, the number of honey bees decreases every year because of the effects of climate change, environmental pollution, and so on. As a result, finding a useful solution to this problem has been more and more attracting scientists and companies. Applying machine learning (ML) methods based on audio data recording inside the hive is a promising solution to detect changes in the beehive. In this study, we investigate the genetic programming (GP) method, one of the powerful ML methods, for identifying bee sound data. We also compare our proposal with the results from a previous study. The experiment results show that with the right configuration of parameters, GP can achieve better results than well-known methods for the task of classifying bee sound samples.

Multi-objective optimization evolutionary algorithms (MOEAs) is one of the most well-known approaches for solving the multi-objective optimization problems (MOPs). When the number of objectives is greater than three, the MOPs are considered as many-objective optimization problem (MaOPs), and many-objective optimization evolutionary algorithms (MaOEAs) are proposed to solve MaOPs. However, MaOPs often contain redundant objectives which do not conflict to any other objectives or even correlate positively to some other. These redundant objectives seriously degrade the efficiency of MOEAs/MaOEAs, so they should be removed to help MaOEAs work better. Therefore, this paper proposes new objective reduction algorithms which uses a Pareto corner search algorithm (PCSEA) to generate non-dominated solutions at corners of Pareto front (PF), and then applies machine learning techniques to remove redundant objectives. The proposed methods not only promote the strengths of PCSEA in finding non-dominated solutions but also promote the strengths of machine learning algorithms in automatically finding the optimal set of objectives. The experiments on 36 instances of DTLZ5(I,M) and 10 instances of WFG3(M) show that the proposed methods can more often find the right set of objectives than six other benchmark methods, respectively about 80% success compared to about 62% success of the other methods.

Gene expression data has been successfully used for cancer classification. However, gene expression data often suffers from a large number of missing values which makes serious issues for classification. A common approach to performing classification with incomplete data is to use imputation methods for estimating missing values before constructing classifiers. However, due to a large number of redundant features, imputation methods for gene expression data are ineffective and inefficient. Feature selection is a popular way to remove redundant features, but it has not been investigated to improve imputation for gene expression data. Therefore, this paper proposes an integration feature selection with imputation to solve the problem. Experimental results show that the proposed method not only improves the classification accuracy, but also speed up the imputation process.

The expression levels of genes is well-recognised to hold the keys to address many fundamental biological problems. A major application of such datasets is cancer diagnosis which is essentially a classification task. Ensemble learning, which is a powerful machine learning approach, has been widely used to improve the performance of many real-world classification problems. Ensemble learning has been also applied for cancer classification from gene expression data. This paper proposed two hybrid ensemble machine learning approaches for classifying cancer gene expression data. The first approach is the integration of random subspace ensemble with bagging, and the second one is the integration of random subspace ensemble with boosting. Experimental results show that the proposed methods can improve classification accuracy for cancer classification from gene expression data.

One of the most challenging issues in the utilisation of machine learning in face datasets is the lack of data, especially when there is inadequate collection of datasets. On one hand, the cost of collecting new face images could be very costly and it depend heavily on the resources and the availability of the data collection. On the other hand, insufficient face datasets could lead to over-fitting issues in any deep learning models especially in the face verification tasks as it requires adequate amount of face dataset. Nevertheless, Generative Adversarial Networks (GANs) offers a better way to augment the data by generating synthetic face images based on the close-distributed pixels of real images. With this intention, GAN inversion was introduced to produce better performance comparing to the previous GAN concepts; by inverting a given face image back into the latent space of a pretrained GAN model with low loss transmissions. This paper demonstrates the feasibility of GAN inversion during the face verification process. We will also illustrate the comparison between previous GAN models, and traditional machine learning augmentation methods in face images generation.

Speaker recognition is the process of identifying an individual from their voices, and it has been widely applied in many real-world applications. Recently, deep learning has instigated a revolutionary high success rate in speaker recognition. The major advantage of deep learning over conventional methods for speaker recognition is attributed to its representation ability, and the ability to produce highly abstract embedding features from utterances. Recent researches had revealed that deep learning method in learning speaker features from raw data, is strongly depending on a speaker's language. However, only minimal researches had done on deep learning over Vietnamese speaker recognition to present. Nevertheless, this paper has proposed a deep transfer learning method which integrates both transfer learning and deep learning to build models for Vietnamese speaker recognition. Our experimental results indicated that the proposed method is able to build accurate models for Vietnamese speaker recognition.

Missing values present as the most common problem in real-world data science. Inadequate treatment of missing values could often result in mass errors. Hence missing values should be managed conscientiously for classification. Generative Adversarial Networks (GANs) have been applied for imputing missing values in most recent years. This paper proposes a multiple imputation method to estimate missing values for classification through the integration of GAN and ensemble learning. Our propose method MIGAN utilises GAN to generate different training observations which are then used to conduct ensemble classifiers for classification with missing data. We conducted our experiments examine MIGAN on various data sets as well as comparing MIGAN with the state-of-the-art imputation methods. The experimental results show significant results, which highlights the accuracy of MIGAN in classifying the missing data.

In this paper, we propose an efficient approach for activity recognition in videos with key frame extraction and deep learning architectures, named KFSENet. First, we propose a key frame selection technique in a motion sequence of 2D frames based on gradient of optical flow to select the most important frames which characterize different actions. From these frames, we extract key points using pose estimation techniques and employ them further in an efficient Deep learning network to learn the action model. In this way, the proposed method be able to remove insignificant frames and decrease the length of the motion vector. We only consider the remaining essential informative frames in the process of action recognition, thus the proposed method is sufficiently fast and robust. We evaluate the proposed method intensively on public dataset named UCF Sport and our self-built HNH dataset in our experiments. We verify that our proposed algorithm receive state-of-the-art on these datasets.

Storm surge is a genuine common fiasco coming from the ocean. Therefore, an exact forecast of surges is a vital assignment to dodge property misfortunes and to decrease a chance caused by tropical storm surge. Genetic Programming (GP) is an evolution-based model learning technique that can simultaneously find the functional form and the numeric coefficients for the model. Therefore, GP has been widely applied to build models for predictive problems. However, GP has seldom been applied to the problem of storm surge forecasting. In this paper, we propose a new method to use GP for evolving models for storm surge forecasting. Experimental results on datasets collected from the Tottori coast of Japan show that GP can evolve accurate storm surge forecasting models. Moreover, GP can automatically select relevant features when evolving storm surge forecasting models, and the models evolved by GP are interpretable.

The current study developed storm surge hindcast/forecast models with lead times of 5, 12, and 24 h at the Sakaiminato port, Tottori, Japan, using the group method of data handling (GMDH) algorithm. For training, local meteorological and hydrodynamic data observed in Sakaiminato during Typhoons Maemi (2003), Songda (2004), and Megi (2004) were collected at six stations. In the forecast experiments, the two typhoons, Maemi and Megi, as well as the typhoon Songda, were used for training and testing, respectively. It was found that the essential input parameters varied with the lead time of the forecasts, and many types of input parameters relevant to training were necessary for near–far forecasting time-series of storm surge levels. In addition, it was seen that the inclusion of the storm surge level at the input layer was critical to the accuracy of the forecast model.

Malicious software, known as malware, has become urgently serious threat for computer security, so automatic mal-ware classification techniques have received increasing attention. In recent years, deep learning (DL) techniques for computer vision have been successfully applied for malware classification by visualizing malware files and then using DL to classify visualized images. Although DL-based classification systems have been proven to be much more accurate than conventional ones, these systems have been shown to be vulnerable to adversarial attacks. However, there has been little research to consider the danger of adversarial attacks to visualized image-based malware classification systems. This paper proposes an adversarial attack method based on the gradient to attack image-based malware classification systems by introducing perturbations on resource section of PE files. The experimental results on the Malimg dataset show that by a small interference, the proposed method can achieve success attack rate when challenging convolutional neural network malware classifiers.

Many-objective optimization problems (MaOPs) have been gained considerable attention for researcher, recently. MaOPs make a number of difficulties for multi-objective optimization evolutionary algorithms (MOEAs) when solving them. Although, there exist a number of many-objective optimization evolutionary algorithms (MaOEAs) for solving MaOPs, they still face difficulties when the number of objectives of MaOPs increases. One common method to reduce or alleviate these difficulties is to use objective dimensionality reduction (or objective reduction for briefly). Moreover, instead of searching the whole of objective space like existing MOEAs or MaOEAs, Pareto Corner Search Evolutionary (PCSEA) concentrates only on some places of objective space, so it decreases time consuming and then speeds up objective reduction. However, PCSEA-based objective reduction needs to specify a threshold to select or remove objectives, which is not straightforward to do. Based on the idea that more conflict two objectives are, more distant two objectives are; in this paper, we introduce a new objective reduction by integrating PCSEA and k-means, DBSCAN clustering algorithms for solving MaOPs which are assumed containing redundant objectives. The experimental results show that the introduced method can reducing redundant objectives better than PCSEA-based objective reduction. The results further strengthen the links between evolutionary computation and machine learning to address optimization problems.

Gene expression data has been successfully used for cancer classification. However, gene expression data often suffers from a large number of missing values which makes serious issues for classification. A common approach to performing classification with incomplete data is to use decision trees which can directly work with missing data. However, decision trees for gene expression data are often inaccurate due to a large number of genes (very high dimensionality) and a small number of samples in gene expression data. Feature selection is a popular way to deal with the problem. Recently, evolutionary computation techniques such as genetic algorithms (GAs) and particle swarm optimisation (PSO) have been widely used for feature selection. Nonetheless, these evolutionary techniques are often unstable and inaccurate when working with high-dimensional gene expression data. Therefore, this paper proposes a new feature selection method which divides the feature space into subspaces multiple times, and then uses evolutionary computation techniques to perform feature selection on these subspaces. Experimental results show that the proposed method not only improves the classification accuracy, but also selects far fewer and more stable features than other common feature selection methods.

Objective: Diabetes is responsible for considerable morbidity, healthcare utilisation and mortality in both developed and developing countries. Currently, methods of treating diabetes are inadequate and costly so prevention becomes an important step in reducing the burden of diabetes and its complications. Electronic health records (EHRs) for each individual or a population have become important tools in understanding developing trends of diseases. Using EHRs to predict the onset of diabetes could improve the quality and efficiency of medical care. In this paper, we apply a wide and deep learning model that combines the strength of a generalised linear model with various features and a deep feed-forward neural network to improve the prediction of the onset of type 2 diabetes mellitus (T2DM). Materials and methods: The proposed method was implemented by training various models into a logistic loss function using a stochastic gradient descent. We applied this model using public hospital record data provided by the Practice Fusion EHRs for the United States population. The dataset consists of de-identified electronic health records for 9948 patients, of which 1904 have been diagnosed with T2DM. Prediction of diabetes in 2012 was based on data obtained from previous years (2009-2011). The imbalance class of the model was handled by Synthetic Minority Oversampling Technique (SMOTE) for each cross-validation training fold to analyse the performance when synthetic examples for the minority class are created. We used SMOTE of 150 and 30 0 percent, in which 300 percent means that three new synthetic instances are created for each minority class instance. This results in the approximated diabetes:non-diabetes distributions in the training set of 1:2 and 1:1, respectively. Results: Our final ensemble model not using SMOTE obtained an accuracy of 84.28%, area under the receiver operating characteristic curve (AUC) of 84.13%, sensitivity of 31.17% and specificity of 96.85%. Using SMOTE of 150 and 300 percent did not improve AUC (83.33% and 82.12%, respectively) but increased sensitivity (49.40% and 71.57%, respectively) with a moderate decrease in specificity (90.16% and 76.59%, respectively). Discussion and conclusions: Our algorithm has further optimised the prediction of diabetes onset using a novel state-of-the-art machine learning algorithm: the wide and deep learning neural network architecture. (C) 2019 Elsevier B.V. All rights reserved.

Many-objective optimisation problems (MaOPs) have recently received a considerable attention from researchers. Due to the large number of objectives, MaOPs bring serious difficulties to existing multi-objective evolutionary algorithms (MOEAs). The major difficulties includes the poor scalability, the high computational cost and the difficulty in visualisation. A number of many-objective evolutionary algorithms (MaOEAs) has been proposed to tackle MaOPs, but existing MaOEAs have still faced with the difficulties when the number of objectives increases. Real-world MaOPs often have redundant objectives that are not only inessential to describe the Pareto-optimal front, but also deteriorate MaOEAs. A common approach to the problem is to use objective dimensionality reduction algorithms to eliminate redundant objectives. By removing redundant objectives, objective reduction algorithms can improve the search efficiency, reduce computational cost, and support for decision making. The performance of an objective dimensionality reduction strongly depends on nondominated solutions generated by MOEAs/MaOEAs. The impact of objective reduction algorithms on MOEAs and vice versa have been widely investigated. However, the impact of objective reduction algorithms on MaOEAs and vice versa have been rarely investigated. This paper studies the interdependence of objective reduction algorithms on MaOEAs. Experimental results show that combining an objective reduction algorithm with an MOEA can only successfully remove redundant objectives when the total number of objectives is small. In contrast, combining the objective reduction algorithm with an MaOEA can successfully remove redundant objectives even when the total number of objectives is large. Experimental results also show that objective reduction algorithms can significantly improve the performance of MaOEAs.

Many-objective optimisation problems (MaOPs) cause serious difficulties for existing multi-objective evolutionary algorithms (MOEAs). One common way to alleviate these difficulties is to use objective dimensionality reduction. Most existing objective reduction methods are time-consuming because they require MOEAs to run numerous generations. Pareto corner search evolutionary algorithm (PCSEA) was proposed in [18] to speed up objective reduction methods by only seeking corner solutions instead of whole solutions. However, the PCSEA-based objective reduction method in [18] needs to predefine a threshold to select objectives which strongly depends on problems and is not straightforward to obtain. This paper proposes a new objective dimensionality reduction method by integrating PCSEA and principal component analysis (PCA). Thanks to combining advantages of PCSEA and PCA, the proposed method not only can be efficient to eliminate redundant objectives, but also not require to define any parameter in advanced. The experimental results also show that the proposed method can perform objective reduction more successfully than the PCSEA-based objective reduction method. The results further strengthen the links between evolutionary computation and machine learning to address optimization problems.

Missing values are an unavoidable issue in many real-world datasets. One of the most popular approaches to classification with incomplete data is to use imputation to replace missing values with plausible values. However, powerful imputation methods are too computationally intensive when applying a classifier to a new unknown instance. This paper proposes new approaches to integrating imputation, clustering and feature selection for classification with incomplete data in order to improve efficiency without loss of accuracy. Clustering is used to reduce the number of instances used by the imputation. Feature selection is used to remove redundant and irrelevant features of training data which greatly reduces the cost of imputation. The paper also investigates the ability of Differential Evolution (DE) to search feature subsets with incomplete data. Results show that the integration of imputation, clustering and feature selection not only improves classification accuracy, but also dramatically reduces the computation time required to estimate missing values when classifying new instances.