Embedded Feature Selection Method Research Articles

Exploring key factors for long-term vessel incident risk prediction

Factor analysis acts a pivotal role in enhancing maritime safety. Most previous studies conduct factor analysis within the framework of incident-related label prediction, where the developed models can be categorized into short-term and long-term prediction models. The long-term models offer a more strategic approach, enabling more proactive risk management, compared to the short-term ones. Nevertheless, few studies have devoted to rigorously identifying the key factors for the long-term prediction and undertaking comprehensive factor analysis. Hence, this study aims to delve into the key factors for predicting the incident risk levels in the subsequent year given a specific datestamp. The majority of candidate factors potentially contributing to the incident risk are collected from vessels’ historical safety performance data spanning up to five years. An improved embedded feature selection method, which integrates Random Forest classifier with a feature filtering process, is proposed to identify key risk-contributing factors from the candidate pool. A dataset with information of 131,652 vessels collected from 2015 to 2023 is utilized for case study. The results demonstrate superior performances of the proposed method in incident prediction and factor interpretability. Comprehensive analysis is conducted upon the key factors, which could help maritime stakeholders formulate management strategies for incident prevention.

Group-feature (Sensor) selection with controlled redundancy using neural networks

In this work, we present a novel embedded feature selection method based on a Multi-layer Perceptron (MLP) network and generalize it for group-feature or sensor selection problems, which can control the level of redundancy among the selected features or groups and it is computationally more efficient than the existing ones in the literature. Additionally, we have generalized the group lasso penalty for feature selection to encompass a mechanism for selecting valuable groups of features while simultaneously maintaining control over redundancy. We establish the monotonicity and convergence of the proposed algorithm, with a smoothed version of the penalty terms, under suitable assumptions. The effectiveness of the proposed method for both feature selection and group feature selection is validated through experimental results on various benchmark datasets. The performance of the proposed methods is compared with some state-of-the-art methods.

A Gaussian process embedded feature selection method based on automatic relevance determination

In Gaussian Process, feature importance is inversely proportional to the corresponding length scale when applying the Automatic Relevance Determination (ARD) structured kernel function. Features can be selected by ranking them according to their importance. Among the ARD-based feature selection methods, no uniform score exists for quantifying the output variation explained by feature subsets. This study proposes two feature selection approaches using two cumulative feature importance scores, one titled derivative decomposition ratio and the other normalized sensitivity, to determine the optimal feature subset. The performance of the approaches is assessed to test if irrelevant features are accurately identified and if the feature rankings are correct. The approaches are applied to identify relevant dimensionless inputs for a hybrid model estimating liquid entrainment fraction in two-phase flow. The results reveal that the proposed methods can identify the optimal feature subset for the hybrid model without significantly worsening its Root Mean Squared Error.

A Generalized Lightweight Intrusion Detection Model With Unified Feature Selection for Internet of Things Networks

ABSTRACTThe applicability of the Internet of Things (IoT) cutting across different domains has resulted in newer “things” acquiring IP connectivity. These things, technically known as IoT devices, are vulnerable to diverse security threats. Consequently, there has been an exponential increase in IoT malware over the past 5 years, and securing IoT devices from such attacks is a pressing concern in the current era. However, the traditional peripheral security measures do not comply with the lightweight security requirements of the IoT ecosystem. Considering this, we propose a lightweight intrusion detection model for IoT networks (LIDM‐IoT) that demonstrates similar efficiency in exposing malicious activities compared with the existing computationally expensive methods. The crux of the proposed model is that it provides efficient attack detection with lower computational requirements in IoT networks. LIDM‐IoT achieves the feat through a novel unified feature selection strategy that unifies filter‐based and embedded feature selection methods. The proposed feature selection strategy reduces the feature space by 94%. Also, we use only the records of a single attack type to build the model using the XGBoost algorithm. We have tested LIDM‐IoT with unseen attack types to ensure its generalized behavior. The results indicate that the proposed model exhibits efficient attack detection, with a reduced feature set, in IoT networks compared with the state‐of‐the‐art models.

Safe dynamic sparse training of modified RBF networks for joint feature selection and classification

Developing an efficient embedded feature selection method for both binary and multiclass classification problems is a fundamental topic to be further studied. In this paper, the conventional radial basis function (RBF) network is modified for joint feature selection and classification. By utilizing the anisotropic Gaussian basis function, an individual weight parameter is assigned to each feature for feature weighting. The L1-regularized loss function is designed by combining the softmax cross entropy loss with L1 regularization terms imposing on feature and output weights. A specialized active-set limited-memory projected Quasi-Newton (SAL-PQN) algorithm is proposed to minimize the L1-regularized loss function, which can optimize all the adjustable parameters and zero out the redundant feature and output weights simultaneously. Furthermore, a safe dynamic pruning strategy is integrated into the SAL-PQN algorithm for dynamic sparse training of the modified RBF network, which continuously excludes the dynamically zeroed-out feature weights from the training process. It is theoretically assured of yielding the same gradient updates as SAL-PQN, thus the relevant computation of the zeroed-out feature weights in the forward and backward propagations can be safely eliminated. The experimental results demonstrate the effectiveness and efficiency of the proposed methods for joint feature selection and classification, and illustrate the practical utility of the inherent feature weighting capability.

Fuzzy Neighborhood-Based Manifold Learning and Feature Weight Matrix for Multilabel Feature Selection

In recent years, it has been difficult to determine neighborhood radii with almost all feature selection methodologies, and the existing multilabel feature selection approaches neglect these correlations among labels, which results in poor classification efficiency. To address these flaws, a fuzzy neighborhood-based manifold learning model and a feature weight matrix are developed to study multilabel feature selection in this article. First, the difference coefficient of the given samples and the new classification margin of each sample are defined to construct an adaptive fuzzy neighborhood radius. While considering the influences of features and labels on the similarity relationships of samples, a novel fuzzy neighborhood similarity relationship is developed by combining it with the label similarity approach that uses the Jaccard similarity coefficient. Second, the least-squares regression model is introduced to design a loss function for learning the mapping relationship between the feature and label spaces. To constrain the objective function, a regularization term based on the fuzzy neighborhood similarity relationship matrix is developed to establish a manifold learning model. To reduce the complexity of the loss function and to reduce the induced structural risk, the final objective function is constructed by incorporating two graph Laplacian matrices. Third, positive and negative confidence values are presented to incorporate the frequency parameter and to form the dependencies between labels into a new label correlation matrix. A new feature weight matrix integrated with the dependencies between labels is defined to learn the correlations between features and labels. Finally, an embedded feature selection method with an objective function is designed, where the objective function is optimized by the gradient descent scheme. Experiments conducted on 12 multilabel datasets reveal that the proposed approach yields better classification results than do other algorithms.

A method to assist designers in optimizing the exterior styling of vehicles based on key features

Aiming to assist the design and management personnel of automotive products to make correct judgments and give optimization solutions for the design of exterior styling, and to solve the defects of the traditional method which relies too much on experience and personal ability, a neural network-based key feature analysis and optimization method of styling is proposed. Firstly, the user attributes are analyzed and quantified, and the weighted optimization of user evaluation data is completed based on the multidimensional Gaussian distribution function to construct a more objective car styling evaluation dataset. Secondly, a particle swarm algorithm is proposed to optimize the prediction model of radial basis function neural network (PSO-RBFNN) which deeply analyzes the mapping relationship between different car perspectives and users’ perceptual preferences. The identification of styling components is completed with the annotation of feature points, and a parametric expression method of styling feature data with multi-viewpoint weight optimization is proposed. Third, the model interpretation is based on the embedded feature selection method to discover the key feature points affecting user perception, and the key common features of styling are obtained through the coupling relationship analysis of feature points. Finally, the optimal distribution range of key features is given based on statistical analysis, and the application process is shown by an example of optimization to verify the feasibility of the theory. Compared with the traditional methods, PSO-RBFNN has better prediction accuracy. Meanwhile, the modeling optimization method based on key features has the great advantages of objective accuracy and high efficiency compared with the traditional theory.

A feature selection method for multimodal multispectral LiDAR sensing

Optical remote sensing techniques can indicate the properties of objects by observing different modalities (physical quantities) of the backscattered light at different optical wavelengths. Established examples are reflectance, fluorescence, Raman, or depolarization spectroscopy. LiDAR sensing, on the other hand, allows acquiring the geometry of objects by measuring the propagation delay of optical probing signals. Multimodal multispectral (MM) LiDAR combines these capabilities and extends conventional monochromatic LiDAR in both spectral and modal dimensions within a single instrument, thus enriching point cloud data with non-geometric information. The potentially high dimension of MM LiDAR data, however, poses significant challenges for instrumental design, data acquisition, and data processing. MM LiDAR data are structured as several or all modalities are available in each of the spectral channels. The above challenges can thus be mitigated by feature selection (FS), if the structure of the features is taken into account, i.e., if entire spectral channels or entire modalities are selected or omitted. Herein, we focus on the feature selection method for MM LiDAR and propose a multiclass group feature selection algorithm (MGSVM FS) consisting of a structural sparsity-based embedded feature selection method with an all-in-one support vector machine (SVM). It tackles jointly the challenges arising from the high dimension of the MM data and the need for a multiclass classification task while exploiting the structure of the MM data. In addition, we introduce a complete workflow for evaluating the feature selection and for decision-making. We apply the framework to selecting an optimum spectral and modal configuration for remote material classification using an experimental MM LiDAR system that provides reflectance, distance, and degree of linear polarization in 28 spectral channels of 10 nm width. For the experimental investigation, we use MM LiDAR data obtained in a controlled lab environment from thirty specimens of four material classes relevant for construction. Using all three modalities, we find a configuration with only 3 spectral channels that achieves a classification mean-F1 score of 100% within this small dataset. Similar classification performance can also be achieved with only two modalities when using more spectral channels. MGSVM FS improves the classification mean-F1 score by up to 25% as compared to random selection and outperforms two other commonly used filter and embedded feature selection methods, in this application example. The proposed group feature selection algorithm and decision-making are useful for MM LiDAR, providing a link between instrumental design, data acquisition, and data processing. However, they are also transferable to other application fields related to multiclass classification, regression, and knowledge discovery, with features structured in groups. The collected MM feature dataset, the MGSVM FS algorithm, and the evaluation pipeline are accessible online.11https://github.com/yuhan-yhyh/Dataset_Code_MGSVM-FS-MM-LiDAR.git.

Investigating the potential of EMA-embedded feature selection method for ESVR and LSTM to enhance the robustness of monthly streamflow forecasting from local meteorological information

Accurate forecast of monthly streamflow is helpful to improve the social capability in risk management. Diverse input features are crucial to the accuracy of machine learning-based monthly streamflow forecast models. The embedded feature selection (EFS) method based on binary-coded metaheuristic algorithm is adept at parallel optimization of feature selection and model hyperparameter. It has been rarely studied coupled with deep learning models in monthly streamflow forecasting. This study further investigated the potential of EFS method using the well-known Long short-term memory network (LSTM) and a state-of-the-art model named Extreme support vector regression (ESVR), with the Evolutionary mating algorithm (EMA) as the search algorithm. The benchmark models were established using EMA for hyperparameter optimization without feature selection. Ten basins with distinct hydroclimate characteristics minimally impacted by human activities from the Catchment Attributes and Meteorology for Large-sample Studies (CAMELS) data set were used to comprehensively examine the proposed models. Nonparametric approaches including the Friedman test and Wilcoxon signed ranks test were utilized to rank the models and detect the significance of the performance variation brought by EFS. The kernel ESVR model coupled with EFS-EMA (EFS-KESVREMA) ranked the first in the 1-month-ahead experiment with NSE = 0.734–0.944. EFS-ESVREMA and EFS-LSTMEMA ranked second and third, with NSE = 0.746–0.94 and NSE = 0.701–0.94 respectively. The results also demonstrated that EFS-EMA helps the benchmark models improve prediction accuracy at a confidence level of 95 % but it has no obvious effect on the standard SVR. As the forecast horizon increased to 3- and 5-month-ahead, EFS-ESVREMA maintained a reliable performance. Whereas EFS-LSTMEMA experienced a significant accuracy decrease in the 5-month-ahead experiment. The proposed models and relevant findings in this study have certain reference value for researchers in monthly streamflow forecasting.

A novel hybrid model combined with ensemble embedded feature selection method for estimating reference evapotranspiration in the North China Plain

The reference evapotranspiration (ETo) is a key parameter in achieving sustainable use of agricultural water resources. To accurately acquire ETo under limited conditions, this study combined the northern goshawk optimization algorithm (NGO) with the extreme gradient boosting (XGBoost) model to propose a novel NGO-XGBoost model. The performance of this model was evaluated using meteorological data from 30 stations in the North China Plain and compared with XGBoost, random forest (RF), and k nearest neighbor (KNN) models. An ensemble embedded feature selection (EEFS) method combined with the results from RF, XGBoost, adaptive boosting (AdaBoost), and categorical boosting (CatBoost) models is used to obtain the importance of meteorological factors in estimating ETo, and thereby determine the optimal combination of inputs to the model. The results indicated that by using the top 3, 4, and 5 important factors as input combinations, all models achieved high ETo estimation accuracy. It is worth noting that there were significant spatial differences in the estimation precisions of the four models, but the NGO-XGBoost model exhibited consistently high estimation precisions, with global performance indicator (GPI) rankings of 1st, and the range of coefficient of determination (R2), nash efficiency coefficient (NSE), root mean square error (RMSE), mean absolute error (MAE) and mean bias error (MBE) were 0.920–0.998, 0.902–0.998, 0.078–0.623 mm d−1, 0.058–0.430 mm d−1, and −0.254–0.062 mm d−1, respectively. Furthermore, the accuracy of the NGO-XGBoost model in estimating ETo varied across different seasons, which was more significantly affected by humidity and wind speed in winter. When the target station data was insufficient, the NGO-XGBoost model was trained by using the historical data from neighboring stations and still maintained a high precision. Overall, this study recommends a reliable method for estimating ETo, which provides a reference for accurately calculating ETo in the North China Plain in the absence of meteorological data.

Finding New VEGFR2 Inhibitors Using Support Vector Machine Classification Model

Introduction: In our current era, the prevalence of cancer and its associated mortality rates have become a pressing concern. As such, finding effective methods for treating cancer has become a matter of significant importance. Abnormal angiogenesis is one of the common characteristics of different types of cancer. So far, the inhibition of vascular endothelial growth factor receptor 2 signaling pathway has received much attention due to its pro-angiogenic role. Therefore, finding reliable computational models to identify inhibitors can be effective in reducing time and cost. The purpose of this study was to use the support vector machine method to classify compounds into two inhibitory and non-inhibitory groups.
 Methods: In order to implement the machine learning model, the ligands studied in this research were extracted from the https://www.bindingdb.org database and after passing the necessary pre-processing, some filter-based and embedded feature selection methods were used. After extracting the descriptors from the data, using the feature selection algorithm based on correlation, the dimensions of the data have been reduced in order to avoid overfitting the model. The classification task utilized a support vector machine model, employing various kernels such as Radial Basis Function (RBF), Polynomial, Sigmoid, and Linear.
 Results: The implementation of the support vector machine model with the RBF kernel along with the feature selection method based on correlation has resulted in a higher accuracy of 82.4% (P=0.008) compared to other feature selection methods used in this study.
 Conclusion: Observations indicate that the correlation-based feature selection method is more accurate than other methods used in this study.

Discriminative multi-label feature selection with adaptive graph diffusion

Feature selection can alleviate the problem of the curse of dimensionality by selecting more discriminative features, which plays an important role in multi-label learning. Recently, embedded feature selection methods have received increasing attentions. However, most existing methods learn the low-dimensional embeddings under the guidance of the local structure between the original instance pairs, thereby ignoring the high-order structure between instances and being sensitive to noise in the original features. To address these issues, we propose a feature selection method named discriminative multi-label feature selection with adaptive graph diffusion (MFS-AGD). Specifically, we first construct a graph embedding learning framework equipped with adaptive graph diffusion to uncover a latent subspace that preserves the higher-order structure information between four tuples. Then, the Hilbert–Schmidt independence criterion (HSIC) is incorporated into the embedding learning framework to ensure the maximum dependency between the latent representation and labels. Benefiting from the interactive optimization of the feature selection matrix, latent representation and similarity graph, the selected features can accurately explore the higher-order structural and supervised information of data. By further considering the correlation between labels, MFS-AG is extended to a more discriminative version,i.e., LMFS-AG. Extensive experimental results on various benchmark data sets validate the advantages of the proposed MFS-AGD and LMFS-AGD methods.

An embedded feature selection method based on generalized classifier neural network for cancer classification

The selection of relevant genes plays a vital role in classifying high-dimensional microarray gene expression data. Sparse group Lasso and its variants have been employed for gene selection to capture the interactions of genes within a group. Most of the embedded methods are linear sparse learning models that fail to capture the non-linear interactions. Additionally, very less attention is given to solving multi-class problems. The existing methods create overlapping groups, which further increases dimensionality. The paper proposes a neural network-based embedded feature selection method that can represent the non-linear relationship. In an effort toward an explainable model, a generalized classifier neural network (GCNN) is adopted as the model for the proposed embedded feature selection. GCNN has well-defined architecture in terms of the number of layers and neurons within each layer. Each layer has a distinct functionality, eliminating the obscure nature of most neural networks. The paper proposes a feature selection approach called Weighted GCNN (WGCNN) that embeds feature weighting as a part of training the neural network. Since the gene expression data comprises a large number of features, to avoid overfitting of the model a statistical guided dropout is implemented at the input layer. The proposed method works for binary as well as multi-class classification problems likewise. Experimental validation is carried out on seven microarray datasets on three learning models and compared with six state-of-art methods that are popularly employed for feature selection. The WGCNN performs well in terms of the F1 score and the number of features selected.

Feature selection under budget constraint in medical applications: analysis of penalized empirical risk minimization methods

Feature selection is a crucial step when building supervised predictive models. In many medical applications, features are associated with costs. For example, the diagnostic value extracted by a clinical test is associated with its own cost. Costs can also refer to a non-financial aspects, such as a decision between an invasive exploratory surgery and a simple blood test. Traditional feature selection methods, which ignore costs, aim to choose a subset of features that maximize the accuracy of the corresponding model. However, such a model can be impractical as the total cost of making a prediction may exceed the assumed user-specified budget. In cost-constrained methods, it is necessary to take into account both the relevance of the feature and its cost. We focus on embedded feature selection methods based on a very general penalized empirical risk minimization framework that includes various loss functions. The most natural ℓ0\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\ell _0$$\\end{document}-type penalty is computationally intractable and therefore we analyze other penalties such as: the cost-sensitive lasso and adaptive lasso, non-convex penalties and the method based on knockoffs. The experiments performed on real medical datasets, including large database MIMIC, indicate that non-convex penalties give promising results, in particular they allow to achieve high accuracy, especially when the assumed budget is low. Our model achieved AUC 0.88 for the MIMIC-II dataset in which we predict the occurrence of liver diseases based on clinical features with a budget equal to (5%) of the cost of all available features, which is significantly better than the AUC for the traditional method. Moreover, the fraction of feature cost wasted for noisy features in our method is usually lower than for cost-sensitive lasso.

GradWise: A Novel Application of a Rank-Based Weighted Hybrid Filter and Embedded Feature Selection Method for Glioma Grading with Clinical and Molecular Characteristics.

Glioma grading plays a pivotal role in guiding treatment decisions, predicting patient outcomes, facilitating clinical trial participation and research, and tailoring treatment strategies. Current glioma grading in the clinic is based on tissue acquired at the time of resection, with tumor aggressiveness assessed from tumor morphology and molecular features. The increased emphasis on molecular characteristics as a guide for management and prognosis estimation underscores is driven by the need for accurate and standardized grading systems that integrate molecular and clinical information in the grading process and carry the expectation of the exposure of molecular markers that go beyond prognosis to increase understanding of tumor biology as a means of identifying druggable targets. In this study, we introduce a novel application (GradWise) that combines rank-based weighted hybrid filter (i.e., mRMR) and embedded (i.e., LASSO) feature selection methods to enhance the performance of feature selection and machine learning models for glioma grading using both clinical and molecular predictors. We utilized publicly available TCGA from the UCI ML Repository and CGGA datasets to identify the most effective scheme that allows for the selection of the minimum number of features with their names. Two popular feature selection methods with a rank-based weighting procedure were employed to conduct comprehensive experiments with the five supervised models. The computational results demonstrate that our proposed method achieves an accuracy rate of 87.007% with 13 features and an accuracy rate of 80.412% with five features on the TCGA and CGGA datasets, respectively. We also obtained four shared biomarkers for the glioma grading that emerged in both datasets and can be employed with transferable value to other datasets and data-based outcome analyses. These findings are a significant step toward highlighting the effectiveness of our approach by offering pioneering results with novel markers with prospects for understanding and targeting the biologic mechanisms of glioma progression to improve patient outcomes.

An embedded feature selection approach for depression classification using short text sequences

Depression has become a serious mental health issue worldwide, particularly due to the rise of the Global Pandemic. Identifying depression of an individual from short texts shared in social media is a challenging task. The present work aims to select the optimal feature subset for classifying short texts for depression detection. By performing feature selection, it is possible to eliminate redundant and noisy features in high-dimensional datasets with small sample sizes. This can prevent the ”curse of dimensionality” and enhance the effectiveness of classification algorithms. However, current feature selection methods often focus on optimizing classification or clustering performance, while neglecting the stability of the selected features. This can lead to unstable results and make it challenging to identify meaningful and interpretable features. This paper introduces a novel embedded feature selection approach named Statistical Relevance Class Frequency based on Whale Optimization Algorithm (SRCF-WOA) for selecting feature subsets from short texts in social media. The proposed methodology extracts both the unigram features and composite features to capture the semantic and structural information. χ2.rcf (Chi-squared relevance class frequency) filter approach is applied to rank the extracted features to signify the importance of the features. WOA is adapted to retrieve the optimal subset of features with low-dimensional space using its high exploration and high exploitation capability. In the evaluation process, four benchmark short text datasets and two classifiers are used. The comparison shows that the proposed embedded feature selection method outperforms other algorithms in terms of accuracy and Fβ scores(β=0.5,1, and 2). The sensitivity analysis is carried out to check the robustness and stability of the proposed method. The findings indicate that the SRCF-WOA surpasses other methods on the majority of datasets, achieving the maximum classification accuracy while utilizing the minimal features. The statistical importance of these findings is further supported by the Analysis of Variance (ANOVA) F-test. Moreover, the proposed method strikes the optimal balance between classification accuracy and feature stability.

Enhancing the prediction of IDC breast cancer staging from gene expression profiles using hybrid feature selection methods and deep learning architecture.

Prediction of the stage of cancer plays an important role in planning the course of treatment and has been largely reliant on imaging tools which do not capture molecular events that cause cancer progression. Gene-expression data-based analyses are able to identify these events, allowing RNA-sequence and microarray cancer data to be used for cancer analyses. Breast cancer is the most common cancer worldwide, and is classified into four stages - stages 1, 2, 3, and 4 [2]. While machine learning models have previously been explored to perform stage classification with limited success, multi-class stage classification has not had significant progress. There is a need for improved multi-class classification models, such as by investigating deep learning models. Gene-expression-based cancer data is characterised by the small size of available datasets, class imbalance, and high dimensionality. Class balancing methods must be applied to the dataset. Since all the genes are not necessary for stage prediction, retaining only the necessary genes can improve classification accuracy. The breast cancer samples are to be classified into 4 classes of stages 1 to 4. Invasive ductal carcinoma breast cancer samples are obtained from The Cancer Genome Atlas (TCGA) and Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) datasets and combined. Two class balancing techniques are explored, synthetic minority oversampling technique (SMOTE) and SMOTE followed by random undersampling. A hybrid feature selection pipeline is proposed, with three pipelines explored involving combinations of filter and embedded feature selection methods: Pipeline 1 - minimum-redundancy maximum-relevancy (mRMR) and correlation feature selection (CFS), Pipeline 2 - mRMR, mutual information (MI) and CFS, and Pipeline 3 - mRMR and support vector machine-recursive feature elimination (SVM-RFE). The classification is done using deep learning models, namely deep neural network, convolutional neural network, recurrent neural network, a modified deep neural network, and an AutoKeras generated model. Classification performance post class-balancing and various feature selection techniques show marked improvement over classification prior to feature selection. The best multiclass classification was found to be by a deep neural network post SMOTE and random undersampling, and feature selection using mRMR and recursive feature elimination, with a Cohen-Kappa score of 0.303 and a classification accuracy of 53.1%. For binary classification into early and late-stage cancer, the best performance is obtained by a modified deep neural network (DNN) post SMOTE and random undersampling, and feature selection using mRMR and recursive feature elimination, with an accuracy of 81.0% and a Cohen-Kappa score (CKS) of 0.280. This pipeline also showed improved multiclass classification performance on neuroblastoma cancer data, with a best area under the receiver operating characteristic (auROC) curve score of 0.872, as compared to 0.71 obtained in previous work, an improvement of 22.81%. The results and analysis reveal that feature selection techniques play a vital role in gene-expression data-based classification, and the proposed hybrid feature selection pipeline improves classification performance. Multi-class classification is possible using deep learning models, though further improvement particularly in late-stage classification is necessary and should be explored further.

Double groups of gates based Takagi-Sugeno-Kang (DG-TSK) fuzzy system for simultaneous feature selection and rule extraction

Embedded feature selection methods based on fuzzy model have been well studied, in which the gate functions are usually used to distinguish the essential features. Fuzzy rules play a critical role in constructing fuzzy systems. However, few works simultaneously evaluate the quality of fuzzy rules during the fuzzy system based feature selection approaches. This article proposes double groups of gates based Takagi-Sugeno-Kang (DG-TSK) fuzzy model to simultaneously conduct feature selection and rule extraction in one training phase. Firstly, a novel gate function is introduced, which is embedded into the system. We call this function as M-gate since it is shaped like an M. Secondly, a new type of fuzzy rule base (FRB) is designed by transforming each data point into one fuzzy rule, which is named point-based FRB (P-FRB). Based on the M-gate and P-FRB, DG-TSK model is developed, in which M-gates are embedded in the antecedents for feature selection and consequents for rule extraction. Note that these two procedures are performed at the same time. We test DG-TSK on 18 classification datasets and compare it with other similar methods. The experiment results demonstrate that DG-TSK is effective.

Automated diagnosis of Retinopathy of prematurity from retinal images of preterm infants using hybrid deep learning techniques

SignificanceRetinopathy of prematurity (ROP) is a retinal disease in premature infants, which, when left untreated, can cause permanent blindness. Due to high variability and interobserver inconsistency in the diagnosis of ROP, there is a significant need to develop an automated system for the prediction of ROP. Although several methods have been used in automated diagnosis in ROP, no dedicated models have been described with adequate performance. Objectivesi) To develop a hybrid deep learning network for the prediction of ROP that can be used in the mass screening of infants. ii) To compare the performance of the proposed hybrid model with the Machine Learning classifiers and pre-trained CNN model. MethodsThe hybrid network is trained with 3200 and tested with 800 infant fundus images. Modified MultiResUNet and matched filter with first-order Gaussian derivative are used to segment the retinal vessels from the fundus images. The Gray level co-occurrence matrix (GLCM) and contour features of segmented images are extracted and selected using an embedded feature selection method. The selected features are evaluated using permutation importance and classified using the Random Forest classifier. ResultsThe proposed hybrid model predicts ROP with an accuracy of 94.5%, sensitivity of 94%, and specificity of 93%, surpassing the machine learning classifiers and the pre-trained models with reduced parameters and being computationally inexpensive. ConclusionsThe proposed hybrid model improves the quality of infant care by providing additional diagnostic assistance to clinicians. It also enhances the accessibility to remote healthcare centers for large-scale automated screening systems.

Research on the Prediction of Operator Users’ Number Portability Based on Community Detection

In 2019, China introduced a policy on Number Portability Management, which has resulted in a rapid increase in the number of lost users among telecom companies. Telecom companies must urgently distinguish those with a tendency toward number portability. However, existing prediction research lacks the input of temporal variations in user data and the graph-based analysis of user relationship characteristics, resulting in a poor prediction effect. In this paper, a neural-network-based approach has been applied to address the limitation, whereby user data do not feature temporal variation. Furthermore, innovative approaches have been proposed to construct multilayer community networks through users’ geographic attributes and to analyze community networks with a network embedding method based on the matrix factorization framework. This fills a gap in existing research areas, whereby the geographic attributes of users have not received much attention. Considering the extensive inputs and multiple features of the predicted attributes, in this paper, the strengths and weaknesses of three feature selection methods are compared, as well as the prediction accuracy of each of the five prediction models. Finally, the embedded feature selection method, deep neural network model, and the Light GBM model are shown to provide better results. After introducing the user community network, it was found that the prediction evaluation indicators of both the deep neural network model and the Light GBM model are improved.