Missing Feature Values Research Articles

ComBat models for harmonization of resting-state EEG features in multisite studies

ObjectivePooling multisite resting-state electroencephalography (rsEEG) datasets may introduce bias due to batch effects (i.e., cross-site differences in the rsEEG related to scanner/sample characteristics). The Combining Batches (ComBat) models, introduced for microarray expression and adapted for neuroimaging, can control for batch effects while preserving the variability of biological covariates. We aim to evaluate four ComBat harmonization methods in a pooled sample from five independent rsEEG datasets of young and old adults. MethodsRsEEG signals (n = 374) were automatically preprocessed. Oscillatory and aperiodic rsEEG features were extracted in sensor space. Features were harmonized using neuroCombat (standard ComBat used in neuroimaging), neuroHarmonize (variant with nonlinear adjustment of covariates), OPNested-GMM (variant based on Gaussian Mixture Models to fit bimodal feature distributions), and HarmonizR (variant based on resampling to handle missing feature values). Relationships between rsEEG features and age were explored before and after harmonizing batch effects. ResultsBatch effects were identified in rsEEG features. All ComBat methods reduced batch effects and features’ dispersion; HarmonizR and OPNested-GMM ComBat achieved the greatest performance. Harmonized Beta power, individual Alpha peak frequency, Aperiodic exponent, and offset in posterior electrodes showed significant relations with age. All ComBat models maintained the direction of observed relationships while increasing the effect size. ConclusionsComBat models, particularly HarmonizeR and OPNested-GMM ComBat, effectively control for batch effects in rsEEG spectral features. SignificanceThis workflow can be used in multisite studies to harmonize batch effects in sensor-space rsEEG spectral features while preserving biological associations.

Processing and optimized learning for improved classification of categorical plant disease datasets

PURPOSE: Crop diseases can cause significant reductions in yield, subsequently impacting a country’s economy. The current research is concentrated on detecting diseases in three specific crops – tomatoes, soybeans, and mushrooms, using a real-time dataset collected for tomatoes and two publicly accessible datasets for the other crops. The primary emphasis is on employing datasets with exclusively categorical attributes, which poses a notable challenge to the research community. METHODS: After applying label encoding to the attributes, the datasets undergo four distinct preprocessing techniques to address missing values. Following this, the SMOTE-N technique is employed to tackle class imbalance. Subsequently, the pre-processed datasets are subjected to classification using three ensemble methods: bagging, boosting, and voting. To further refine the classification process, the metaheuristic Ant Lion Optimizer (ALO) is utilized for hyper-parameter tuning. RESULTS: This comprehensive approach results in the evaluation of twelve distinct models. The top two performers are then subjected to further validation using ten standard categorical datasets. The findings demonstrate that the hybrid model II-SN-OXGB, surpasses all other models as well as the current state-of-the-art in terms of classification accuracy across all thirteen categorical datasets. II utilizes the Random Forest classifier to iteratively impute missing feature values, employing a nearest features strategy. Meanwhile, SMOTE-N (SN) serves as an oversampling technique particularly for categorical attributes, again utilizing nearest neighbors. Optimized (using ALO) Xtreme Gradient Boosting OXGB, sequentially trains multiple decision trees, with each tree correcting errors from its predecessor. CONCLUSION: Consequently, the model II-SN-OXGB emerges as the optimal choice for addressing classification challenges in categorical datasets. Applying the II-SN-OXGB model to crop datasets can significantly enhance disease detection which in turn, enables the farmers to take timely and appropriate measures to prevent yield losses and mitigate the economic impact of crop diseases.

Rough topological structures by various types of maximal neighborhoods

<p>This manuscript centers on creating various topologies utilizing different sorts of maximal neighborhoods. The comparison of these topologies with the previous ones reveal that the earlier topology is weaker than the current ones. The core properties of the proposed topologies are examined, and the necessary conditions for achieving certain equivalences among them are outlined. Additionally, this study provides a distinctive characterization of these topologies by pinpointing the coarsest and largest one among all types, whereas previous methods were limited to characterizing only disjoint pairs of sets. Thereafter, these topologies are utilized to evolve new approximations. One of the major benefits of the current extension is that it adheres to all the properties of the original approximations without the constraints or limitations imposed by earlier versions. The significance of this paper lies not only in introducing new types of approximations based primarily on different kinds of topologies, but also in the fact that these approximations maintain the monotonic property for any given relation, enabling effective evaluation of uncertainty in the data. The monotonic property is crucial for various applications, as it guarantees that the approximation process is logically coherent and robust in the face of evolving information. The proposed models distinguish from their predecessors by their ability to compare all types of the suggested approximations. Moreover, comparisons reveal that the optimal approximations and accuracy are achieved with a specific type of generating topologies. The results demonstrate that topological notions can be a potent technique for studying rough set models. Furthermore, advanced topological features of approximate sets aid in finding rough measures, which assists in identifying missing feature values. Afterward, a numerical example is presented to highlight and emphasize the importance of the present results. Ultimately, the benefits of the followed manner are scrutinized and also some of their limitations are pointed out.</p>

Data Preprocessing and Machine Learning Modeling for Rockburst Assessment

Rockbursts pose a significant threat to human safety and environmental stability. This paper aims to predict rockburst intensity using a machine learning model. A dataset containing 344 rockburst cases was collected, with eight inducing features as input and four rockburst grades as output. In the preprocessing stage, missing feature values were estimated using a regression imputation strategy. A novel approach, which combines feature selection (FS), t-distributed stochastic neighbor embedding (t-SNE), and Gaussian mixture model (GMM) clustering, was proposed to relabel the dataset. The effectiveness of this approach was compared with common statistical methods, and its underlying principles were analyzed. A voting ensemble strategy was used to build the machine learning model, and optimal hyperparameters were determined using the tree-structured Parzen estimator (TPE), whose efficiency and accuracy were compared with three common optimization algorithms. The best combination model was determined using performance evaluation and subsequently applied to practical rockburst prediction. Finally, feature sensitivity was studied using a relative importance analysis. The results indicate that the FS + t-SNE + GMM approach stands out as the optimum data preprocessing method, significantly improving the prediction accuracy and generalization ability of the model. TPE is the most effective optimization algorithm, characterized simultaneously by both high search capability and efficiency. Moreover, the elastic energy index Wet, the maximum circumferential stress of surrounding rock σθ, and the uniaxial compression strength of rock σc were identified as relatively important features in the rockburst prediction model.

Prediction of Diabetic Macular Edema Using Knowledge Graph.

Diabetic macular edema (DME) is a significant complication of diabetes that impacts the eye and is a primary contributor to vision loss in individuals with diabetes. Early control of the related risk factors is crucial to reduce the incidence of DME. Artificial intelligence (AI) clinical decision-making tools can construct disease prediction models to aid in the clinical screening of the high-risk population for early disease intervention. However, conventional machine learning and data mining techniques have limitations in predicting diseases when dealing with missing feature values. To solve this problem, a knowledge graph displays the connection relationships of multi-source and multi-domain data in the form of a semantic network to enable cross-domain modeling and queries. This approach can facilitate the personalized prediction of diseases using any number of known feature data. In this study, we proposed an improved correlation enhancement algorithm based on knowledge graph reasoning to comprehensively evaluate the factors that influence DME to achieve disease prediction. We constructed a knowledge graph based on Neo4j by preprocessing the collected clinical data and analyzing the statistical rules. Based on reasoning using the statistical rules of the knowledge graph, we used the correlation enhancement coefficient and generalized closeness degree method to enhance the model. Meanwhile, we analyzed and verified these models' results using link prediction evaluation indicators. The disease prediction model proposed in this study achieved a precision rate of 86.21%, which is more accurate and efficient in predicting DME. Furthermore, the clinical decision support system developed using this model can facilitate personalized disease risk prediction, making it convenient for the clinical screening of a high-risk population and early disease intervention.

Feature selection for multiset-valued data based on fuzzy conditional information entropy using iterative model and matrix operation

Multiset-valued data is a powerful tool to deal with missing feature values. Classical rough set theory is sensitive to noise in classification learning due to the stringent condition of equivalence relation. Thus, a class of fuzzy relations can be introduced to describe the similarity between samples in multiset-valued data. However, these fuzzy relations have deficiencies when they are used in the computation of fuzzy conditional information entropy. This paper proposes a new model for multiset-valued data by introducing two variable parameters: one controls the similarity between samples, the other dominates the distance between feature values. This model employs the iterative computation strategy to define fuzzy conditional information entropy that is computing by matrix. The notions of three fuzzy information entropies in a multiset-valued decision information system (MVDIS) are put forward. For measuring any added feature, the significance based on fuzzy conditional information entropy is introduced. Moreover, fuzzy conditional information entropy iterative model (FCIEI-model) is presented. Finally, feature selection in an MVDIS based on FCIEI-model and matrix operation is researched, and the relevant algorithm (denoted as FCIEIM-MO algorithm) is proposed. The experimental results demonstrate that FCIEIM-MO algorithm possesses good robustness and is superior to some feature selection algorithms. The experimental results demonstrate that FCIEIM-MO algorithm possesses good robustness and exhibit better performance compared with some feature selection algorithms.

Reconstruction of 0.05° all-sky daily maximum air temperature across Eurasia for 2003–2018 with multi-source satellite data and machine learning models

The Eurasian continent is highly vulnerable to climate change. However, there is a lack of high spatiotemporal resolution temperature datasets to support climate change analysis in this region. In this study, an all-sky daily maximum air temperature (Tmax) product at 0.05° spatial resolution across Eurasia for 2003–2018 was developed. This product is generated using a satellite-derived model, including parameters such as daytime and nighttime land surface temperature (LST), downward shortwave radiation, net radiation, leaf area index, enhanced vegetation index, and albedo. The study area, Eurasia, was divided into seven regions using a data-driven method. Four machine learning methods, histogram-based gradient boosting (HGB), extremely randomized trees (ET), random forest (RF), and deep belief network (DBN) were employed to train Tmax estimation models using 4476 stations from GHCN, GSOD, and CMDC. HGB was finally selected since it exhibited the highest estimation accuracy, the determination coefficient (R2) and root-mean-square-error (RMSE) of the HGB model are 0.984 and 1.736 °C with non-missing values in datasets and 0.985 and 1.812 °C with missing values respectively. The impact of the different situations of LST features on HGB models was tested. LST features containing interpolated LST and GLDAS Ta are considered the best. The spatial and temporal accuracy of HGB models was then examined across different land cover types, latitudes, elevations, and months. The permutation test was employed to examine the contributions of different features. Finally, HGB models trained using the best LST features were used to generate the Tmax products. In comparison with existing temperature products, the R2 and RMSE values were reported as 0.980 and 2.177 °C, indicating strong competition among existing Tmax products. In summary, our study provides a scheme for estimating parameters with missing feature values in a consistent manner and provides a solid foundation for the environmental and climate changes studies.

Demand prediction of emergency materials using case-based reasoning extended by the Dempster-Shafer theory

In recent years, the frequent occurrence of natural hazards has caused huge economic and human losses, as well as seriously impacting the sustainable development of society. The effective management of emergency responses to natural hazards has become an important research topic worldwide. The demand prediction of emergency materials is the premise and basis for the optimal allocation of emergency resources, which is of great significance in improving the efficiency of disaster-related emergency responses. Using case-based reasoning (CBR) and the Dempster-Shafer theory, we investigated methods of predicting emergency materials demand. First, to address the problems of missing feature values, feature heterogeneity and inter-correlations among features of CBR, we proposed a case retrieval strategy based on Dempster-Shafer theory that not only lays a theoretical foundation for subsequent research, but also improves the case retrieval strategy used in CBR. Second, inspired by the 4R principle in CBR, we proposed a scenario-matching method for natural hazard, which uses historical cases in the absence of effective decision data for natural hazard-related loss predictions. Third, assuming that the impact of natural hazards will change with time, we further constructed a dynamic prediction model of emergency material demand based on the prediction results of natural hazard losses. In this paper, typhoon and earthquake disasters are used as case studies to demonstrate the application of the proposed materials demand prediction model, and the effectiveness of the method is demonstrated through empirical analysis.

Two‐stage‐neighborhood‐based multilabel classification for incomplete data with missing labels

In recent years, it has been difficult for multilabel classification to obtain complete multilabel data in real-world applications, and even a large number of labels for training samples are randomly missed. As a result, the classification task of incomplete multilabel data with missing labels faces formidable challenges. This paper presents a two-stage-neighborhood-based multilabel classification method for incomplete data with missing labels in neighborhood decision systems. First, to solve the problem of selecting the neighborhood radius manually, as well as balancing the samples in the neighborhood, the neighborhood radius based on the feature distribution function is defined, and the differences and similarities between samples through the identifiable and indiscernible matrices are, respectively, computed. Then, a restoration method for missing feature values is proposed for use in the first stage. Second, to consider the nonlinear relationship among features, a neighborhood-based fuzzy similarity relationship between samples is investigated based on the Gaussian kernel function. By integrating the fuzzy similarity relationship matrix, label-specific feature matrix, and label correlation matrix, an objective function based on the regression model is presented, the optimal solutions to the label-specific feature and label correlation matrices based on the gradient descent strategy are provided, and a new multilabel classification method with missing labels is developed during the second stage. Finally, two-stage multilabel classification algorithms are designed. Experiments on 18 multilabel data sets demonstrate that our designed algorithms are effective not only for recovering missing feature values, but also for improving the classification performance of data with missing labels.

Clustering with Missing Features: A Density-Based Approach

Density clustering has been widely used in many research disciplines to determine the structure of real-world datasets. Existing density clustering algorithms only work well on complete datasets. In real-world datasets, however, there may be missing feature values due to technical limitations. Many imputation methods used for density clustering cause the aggregation phenomenon. To solve this problem, a two-stage novel density peak clustering approach with missing features is proposed: First, the density peak clustering algorithm is used for the data with complete features, while the labeled core points that can represent the whole data distribution are used to train the classifier. Second, we calculate a symmetrical FWPD distance matrix for incomplete data points, then the incomplete data are imputed by the symmetrical FWPD distance matrix and classified by the classifier. The experimental results show that the proposed approach performs well on both synthetic datasets and real datasets.

A Belief Theory-Based Decision System for Breast Cancer Data

Belief theory involving the use of Shafer's model and source combination rules has found wide use for the development and improvement of classification models. Availability of theoretical constructs for making the best use of available information including the ability to deal with ignorance makes it an attractive option. Representation of ignorance means that we can deal with hard to classify samples by deferring the decision rather than taking an erroneous one. This makes it especially useful in the biomedical field where the cost of making an error is high. In this paper, we develop a belief theory-based classification model and apply it to the Wisconsin breast cancer database. The proposed method is compared with an existing belief theory-based classifier, the conventional support vector machine classifier and other recent approaches using different performance parameters such as error rate, false-positive rate, false negative rate, and confusion matrix. The proposed model is found to outperform the approaches under consideration and is also able to deal with data samples having missing feature values effectively.

Selective ensemble of uncertain extreme learning machine for pattern classification with missing features

Ensemble learning is an effective technique to improve performance and stability compared to single classifiers. This work proposes a selective ensemble classification strategy to handle missing data classification, where an uncertain extreme learning machine with probability constraints is used as individual (or base) classifiers. Then, three selective ensemble frameworks are developed to optimize ensemble margin distributions and aggregate individual classifiers. The first two are robust ensemble frameworks with the proposed loss functions. The third is a sparse ensemble classification framework with the zero-norm regularization, to automatically select the required individual classifiers. Moreover, the majority voting method is applied to produce ensemble classifier for missing data classification. We demonstrate some important properties of the proposed loss functions such as robustness, convexity and Fisher consistency. To verify the validity of the proposed methods for missing data, numerical experiments are implemented on benchmark datasets with missing feature values. In experiments, missing features are first imputed by using expectation maximization algorithm. Numerical experiments are simulated in filled datasets. With different probability lower bounds of classification accuracy, experimental results under different proportion of missing values show that the proposed ensemble methods have better or comparable generalization compared to the traditional methods in handling missing-value data classifications.

Spatial Classification with Limited Observations Based on Physics-Aware Structural Constraint

Spatial classification with limited feature observations has been a challenging problem in machine learning. The problem exists in applications where only a subset of sensors are deployed at certain regions or partial responses are collected in field surveys. Existing research mostly focuses on addressing incomplete or missing data, e.g., data cleaning and imputation, classification models that allow for missing feature values, or modeling missing features as hidden variables and applying the EM algorithm. These methods, however, assume that incomplete feature observations only happen on a small subset of samples, and thus cannot solve problems where the vast majority of samples have missing feature observations. To address this issue, we propose a new approach that incorporates physics-aware structural constraints into the model representation. Our approach assumes that a spatial contextual feature is observed for all sample locations and establishes spatial structural constraint from the spatial contextual feature map. We design efficient algorithms for model parameter learning and class inference. Evaluations on real-world hydrological applications show that our approach significantly outperforms several baseline methods in classification accuracy, and the proposed solution is computationally efficient on a large data volume.

Choosing Clinical Variables for Risk Stratification Post-Acute Coronary Syndrome

Most risk stratification methods use expert opinion to identify a fixed number of clinical variables that have prognostic significance. In this study our goal was to develop improved metrics that utilize a variable number of input parameters. We first used Bootstrap Lasso Regression (BLR) – a Machine Learning method for selecting important variables – to identify a prognostic set of features that identify patients at high risk of death 6-months after presenting with an Acute Coronary Syndrome. Using data derived from the Global Registry of Acute Coronary Events (GRACE) we trained a logistic regression model using these features and evaluated its performance on a development set (N = 43,063) containing patients who have values for all features, and a separate dataset (N = 6,363) that contains patients who have missing feature values. The final model, Ridge Logistic Regression with Variable Inputs (RLRVI), uses imputation to estimate values for missing features. BLR identified 19 features, 8 of which appear in the GRACE score. RLRVI had modest, yet statistically significant, improvement over the standard GRACE score on both datasets. Moreover, for patients who are relatively low-risk (GRACE≤87), RLRVI had an AUC and Hazard Ratio of 0.754 and 6.27, respectively, vs. 0.688 and 2.46 for GRACE, (p < 0.007). RLRVI has improved discriminatory performance on patients who have values for the 8 GRACE features plus any subset of the 11 non-GRACE features. Our results demonstrate that BLR and data imputation can be used to obtain improved risk stratification metrics, particularly for patients who are classified as low risk using traditional methods.

Clustering of Data with Missing Entries using Non-convex Fusion Penalties.

The presence of missing entries in data often creates challenges for pattern recognition algorithms. Traditional algorithms for clustering data assume that all the feature values are known for every data point. We propose a method to cluster data in the presence of missing information. Unlike conventional clustering techniques where every feature is known for each point, our algorithm can handle cases where a few feature values are unknown for every point. For this more challenging problem, we provide theoretical guarantees for clustering using a fusion penalty based optimization problem. Furthermore, we propose an algorithm to solve a relaxation of this problem using saturating non-convex fusion penalties. It is observed that this algorithm produces solutions that degrade gradually with an increase in the fraction of missing feature values. We demonstrate the utility of the proposed method using a simulated dataset, the Wine dataset and also an under-sampled cardiac MRI dataset. It is shown that the proposed method is a promising clustering technique for datasets with large fractions of missing entries.

Knowledge Acquisition Approach Based on Incremental Objects From Data With Missing Values

Knowledge acquisition is the process of extracting useful knowledge from data sets to analyze data in areas of data mining and knowledge discovery. Most current knowledge acquisition work mainly focuses on static data. However, due to the dynamic characteristics of data, the objects grow at an unprecedented rate in real-world data sets. The incremental objects with a dynamic environment significantly affect knowledge updating. To maintain the effectiveness of knowledge from the dynamic data, it is necessary to update the knowledge timely. So far, there are relatively few studies on knowledge acquisition for the data with missing feature values, i.e., incomplete data. To handle with this issue, an incremental updating manner of the accuracy matrix and coverage matrix are first proposed on the basis of the computations of the tolerance classes in incomplete data, which plays an important role in the knowledge acquisition process. Then, an incremental knowledge acquisition algorithm is proposed when some new objects added to the data with missing values. Finally, some numerical experiments are conducted to evaluate the efficiency of the proposed algorithm.

Optimal Bayesian Classification With Missing Values

Missing values can be an impediment to designing and applying classifiers. Missing values are common in biomedical studies due to various reasons, including missing tests or complex profiling technologies for different omics measurements in modern biomedicine. Many procedures have been proposed to impute values that are missing. This paper considers missing feature values in the context of optimal Bayesian classification, which selects a classifier that minimizes the expected error with respect to the posterior distribution governing an uncertainty class of feature-label distributions. The missing-value problem fits neatly into the overall framework of optimal Bayesian classification by marginalizing out the missing-value process from the feature-label distribution, and then updating the prior distribution of class-conditional parameters to posterior distributions using new observations. Generally, an optimal Bayesian classifier is defined via the effective class-conditional densities, which are averages of the parameterized feature-label distributions in the uncertainty class relative to the posterior distribution. Hence, once the posterior distribution incorporating the missing value process is found, the optimal Bayesian classifier pertaining to the features with missing values can be derived from the corresponding effective class-conditional densities. This paper presents the general theory, derives a closed-form decision rule for the optimal Bayesian classifier in a Gaussian model with independent features, and utilizes Hamiltonian Monte Carlo for the Gaussian model with arbitrary covariance matrices. Superior performance is demonstrated when compared to linear discriminant analysis, quadratic discriminant analysis, and support vector machines in conjunction with Gibbs sampling imputation using synthetic and real-world omics data.

Ego-network probabilistic graphical model for discovering on-line communities

Community discovery is a leading research topic in social network analysis. In this paper, we present an ego-network probabilistic graphical model (ENPGM) which encodes users’ feature similarities and the causal dependencies between users’ profiles, communities, and ego networks. The model comprises three parts: a profile similarity probabilistic graph, social circle vector, and relationship probabilistic vector. Using Bayesian networks, the profile similarity probabilistic graph considers information about both the features of individuals and network structures with low memory usage. The social circle vector is proposed to describe both the alters belonging to a community and the features causing the community to emerge. The relationship probabilistic vector represents the probability that an ego network forms when given a set of user profiles and a set of circles. We then propose a parameter-learning algorithm and the ego-network probabilistic criterion (ENPC) for extracting communities from ego networks with some missing feature values. The ENPC score balances both the positive and negative impacts of social circles on the probabilities of forming an ego network. Experimental results using Facebook, Twitter, and Google+ datasets indicate that the ENPGM and community learning algorithms can predict social circles with similar quality to the ground-truth communities.

Robust K-Median and K-Means Clustering Algorithms for Incomplete Data

Incomplete data with missing feature values are prevalent in clustering problems. Traditional clustering methods first estimate the missing values by imputation and then apply the classical clustering algorithms for complete data, such as K-median and K-means. However, in practice, it is often hard to obtain accurate estimation of the missing values, which deteriorates the performance of clustering. To enhance the robustness of clustering algorithms, this paper represents the missing values by interval data and introduces the concept of robust cluster objective function. A minimax robust optimization (RO) formulation is presented to provide clustering results, which are insensitive to estimation errors. To solve the proposed RO problem, we propose robust K-median and K-means clustering algorithms with low time and space complexity. Comparisons and analysis of experimental results on both artificially generated and real-world incomplete data sets validate the robustness and effectiveness of the proposed algorithms.

Identification of progressive mild cognitive impairment patients using incomplete longitudinal MRI scans.

Distinguishing progressive mild cognitive impairment (pMCI) from stable mild cognitive impairment (sMCI) is critical for identification of patients who are at risk for Alzheimer's disease (AD), so that early treatment can be administered. In this paper, we propose a pMCI/sMCI classification framework that harnesses information available in longitudinal magnetic resonance imaging (MRI) data, which could be incomplete, to improve diagnostic accuracy. Volumetric features were first extracted from the baseline MRI scan and subsequent scans acquired after 6, 12, and 18 months. Dynamic features were then obtained using the 18th month scan as the reference and computing the ratios of feature differences for the earlier scans. Features that are linearly or non-linearly correlated with diagnostic labels are then selected using two elastic net sparse learning algorithms. Missing feature values due to the incomplete longitudinal data are imputed using a low-rank matrix completion method. Finally, based on the completed feature matrix, we build a multi-kernel support vector machine (mkSVM) to predict the diagnostic label of samples with unknown diagnostic statuses. Our evaluation indicates that a diagnosis accuracy as high as 78.2% can be achieved when information from the longitudinal scans is used-6.6% higher than the case using only the reference time point image. In other words, information provided by the longitudinal history of the disease improves diagnosis accuracy.