Large Number Of Features Research Articles

Machine learning-based feature selection to search stable microbial biomarkers: application to inflammatory bowel disease.

Biomarker discovery exploiting feature importance of machine learning has risen recently in the microbiome landscape with its high predictive performance in several disease states. To have a concrete selection among a high number of features, recursive feature elimination (RFE) has been widely used in the bioinformatics field. However, machine learning-based RFE has factors that decrease the stability of feature selection. In this article, we suggested methods to improve stability while sustaining performance. We exploited the abundance matrices of the gut microbiome (283 taxa at species level and 220 at genus level) to classify between patients with inflammatory bowel disease (IBD) and healthy control (1,569 samples). We found that applying an already published data transformation before RFE improves feature stability significantly. Moreover, we performed an in-depth evaluation of different variants of the data transformation and identify those that demonstrate better improvement in stability while not sacrificing classification performance. To ensure a robust comparison, we evaluated stability using various similarity metrics, distances, the common number of features, and the ability to filter out noise features. We were able to confirm that the mapping by the Bray-Curtis similarity matrix before RFE consistently improves the stability while maintaining good performance. Multilayer perceptron algorithm exhibited the highest performance among 8 different machine learning algorithms when a large number of features (a few hundred) were considered based on the best performance across 100 bootstrapped internal test sets. Conversely, when utilizing only a limited number of biomarkers as a trade-off between optimal performance and method generalizability, the random forest algorithm demonstrated the best performance. Using the optimal pipeline we developed, we identified 14 biomarkers for IBD at the species level and analyzed their roles using Shapley additive explanations. Taken together, our work not only showed how to improve biomarker discovery in the metataxonomic field without sacrificing classification performance but also provided useful insights for future comparative studies.

DATA CLUSTERING BASED ON INDUCTIVE LEARNING OF NEURO-FUZZY NETWORK WITH DISTANCE HASHING

Context. Cluster analysis is widely used to analyze data of various nature and dimensions. However, the known methods of cluster analysis are characterized by low speed and are demanding on computer memory resources due to the need to calculate pairwise distances between instances in a multidimensional feature space. In addition, the results of known methods of cluster analysis are difficult for human perception and analysis with a large number of features.
 Objective. The purpose of the work is to increase the speed of cluster analysis, the interpretability of the resulting partition into clusters, as well as to reduce the requirements of cluster analysis to computer memory.
 Method. A method for cluster analysis of multidimensional data is proposed, which for each instance calculates its hash based on the distance to the conditional center of coordinates, uses a one-dimensional coordinate along the hash axis to determine the distances between instances, considers the resulting hash as a pseudo-output feature, breaking it into intervals, which matches the labels pseudo-classes – clusters, having received a rough crisp partition of the feature space and sample instances, automatically generates a partition of input features into fuzzy terms, determines the rules for referring instances to clusters and, as a result, forms a fuzzy inference system of the Mamdani-Zadeh classifier type, which is further trained in the form of a neuro-fuzzy network to ensure acceptable values of the clustering quality functional. This makes it possible to reduce the number of terms and features used, to evaluate their contribution to making decisions about assigning instances to clusters, to increase the speed of data cluster analysis, and to increase the interpretability of the resulting data splitting into clusters.
 Results. The mathematical support for solving the problem of cluster data analysis in conditions of large data dimensions has been developed. The experiments confirmed the operability of the developed mathematical support have been carried out.
 Conclusions. . The developed method and its software implementation can be recommended for use in practice in the problems of analyzing data of various nature and dimensions.

Optimize temporal configuration for motor imagery-based multiclass performance and its relationship with subject-specific frequency

Enhancing the performance of motor imagery-based Brain-Computer Interfaces (BCI) has been a significant goal in the BCI field. To achieve such a goal, several typical and promising techniques have been implemented, such as developing intelligent algorithms, combining features from different domains, extracting subject-specific parameters, and so forth. Previous studies performing temporal segmentation often ended up with a large number of features and placed a burden on computational cost, which poses a disadvantage to online analysis. This study proposes a novel approach to utilizing short-window segments to find an optimal combination of time segments and feature extractors. Electroencephalogram data from four datasets (BCI Competition IV dataset 2a, 2b and two self-acquired datasets) were segmented into four types of the time window and had features extracted by Common Spatial Pattern and its variants, and lastly classified by Linear Discriminant Analysis. The result shows that the combination of the “2-s with 1-s overlapping” segment and Filter Bank Common Spatial Pattern yields overall accuracy of 2–6.5% (p-value <0.05), higher than other methods in comparison. In addition, the study finds that there is a negative correlation (r = −0.38) between the number of subject-specific frequency bands and the performance (p-value <0.0001). The results demonstrate that the narrower and more focus frequency range chosen, the more accurate the model can achieve. Our results indicate that the “2-s with 1-s overlapping” segment would be an ideal candidate for online BCI analysis, and the response of selected frequency bands could be an informative indicator to evaluate BCI performance.

Bayesian bi-clustering methods with applications in computational biology

Bi-clustering is a useful approach in analyzing large biological data sets when the observations come from heterogeneous groups and have a large number of features. We outline a general Bayesian approach in tackling bi-clustering problems in moderate to high dimensions and propose three Bayesian bi-clustering models on categorical data which increase in complexities in their modeling of the distributions of features across bi-clusters. Our proposed methods apply to a wide range of scenarios: from situations where data are cluster-distinguishable only among a small subset of features but masked by a large amount of noise to situations where different groups of data are identified by different sets of features or data exhibit hierarchical structures. Through simulation studies we show that our methods outperform existing (bi-)clustering methods in both identifying clusters and recovering feature distributional patterns across bi-clusters. We further apply the developed approaches to a human genetic dataset, a human single-cell genomic dataset, and a collection of 1774 mouse genomic datasets with a focus on 58 genes from two pathways.

Hybrid Sequential Feature Selection with Ensemble Boosting Class-based Classification Method

The rapid rise in hacking and computer network assaults throughout the world has highlighted the need for more effective intrusion detection and prevention solutions. The intrusion detection system (IDS) is critical in identifying abnormalities and assaults on the network, which have grown in size and scope. IDS prevents intruders from gaining access to information in the field of network security as a result. The use of IDS is critical for detecting various types of attacks. Because the network traffic dataset contains a large number of features, the process of selecting and removing irrelevant features improves the accuracy of the classification algorithms. For the fact that a large dimension allows us to include more data, the feature vector can be built by combining different types of features. Contains a lot of redundant or irrelevant data can cause confusion. Over-fitting issues and a decrease in the generalization capacity of the model. Solving such a problem necessitates a sequence of feature selection methods the boosted maximum relevance maximum distance (BMRMD) method can report on the contribution of each feature as well as the predictive accuracy based on the best feature set. As a result, the best features in this study were chosen using the BMRMD assesses feature redundancy to determine feature relevance to the target class based on optimum ensemble feature classification

Ensembles of Random SHAPs

The ensemble-based modifications of the well-known SHapley Additive exPlanations (SHAP) method for the local explanation of a black-box model are proposed. The modifications aim to simplify the SHAP which is computationally expensive when there is a large number of features. The main idea behind the proposed modifications is to approximate the SHAP by an ensemble of SHAPs with a smaller number of features. According to the first modification, called the ER-SHAP, several features are randomly selected many times from the feature set, and the Shapley values for the features are computed by means of “small” SHAPs. The explanation results are averaged to obtain the final Shapley values. According to the second modification, called the ERW-SHAP, several points are generated around the explained instance for diversity purposes, and the results of their explanation are combined with weights depending on the distances between the points and the explained instance. The third modification, called the ER-SHAP-RF, uses the random forest for a preliminary explanation of the instances and determines a feature probability distribution which is applied to the selection of the features in the ensemble-based procedure of the ER-SHAP. Many numerical experiments illustrating the proposed modifications demonstrate their efficiency and properties for a local explanation.

Evaluation of Boruta algorithm in DDoS detection

Distributed Denial of Service (DDoS) is a type of attack that leverages many compromised systems or computers, as well as multiple Internet connections, to flood targeted resources simultaneously. A DDoS attack's main purpose is to disrupt website traffic and cause it to crash. As traffic grows over time, detecting a Distributed Denial of Service (DDoS) assault is a challenging task. Furthermore, a dataset containing a large number of features may degrade machine learning’s detection performance. Therefore, in machine learning, it is necessary to prepare a relevant list of features for the training phase in order to obtain good accuracy performance. With far too many possibilities, choosing the relevant feature is complicated. This study proposes the Boruta algorithm as a suitable approach to achieve accuracy in identifying the relevant features. To evaluate the Boruta algorithm, multiple classifiers (J48, random forest, naïve bayes, and multilayer perceptron) were used so as to determine the effectiveness of the features selected by the the Boruta algorithm. The outcomes obtained showed that the random forest classifier had a higher value, with a 100% true positive rate, and 99.993% in the performance measure of accuracy, when compared to other classifiers.

Graph-Based Multi-Label Classification for WiFi Network Traffic Analysis

Network traffic analysis, and specifically anomaly and attack detection, call for sophisticated tools relying on a large number of features. Mathematical modeling is extremely difficult, given the ample variety of traffic patterns and the subtle and varied ways that malicious activity can be carried out in a network. We address this problem by exploiting data-driven modeling and computational intelligence techniques. Sequences of packets captured on the communication medium are considered, along with multi-label metadata. Graph-based modeling of the data are introduced, thus resorting to the powerful GRALG approach based on feature information granulation, identification of a representative alphabet, embedding and genetic optimization. The obtained classifier is evaluated both under accuracy and complexity for two different supervised problems and compared with state-of-the-art algorithms. We show that the proposed preprocessing strategy is able to describe higher level relations between data instances in the input domain, thus allowing the algorithms to suitably reconstruct the structure of the input domain itself. Furthermore, the considered Granular Computing approach is able to extract knowledge on multiple semantic levels, thus effectively describing anomalies as subgraphs-based symbols of the whole network graph, in a specific time interval. Interesting performances can thus be achieved in identifying network traffic patterns, in spite of the complexity of the considered traffic classes.

A malware detection system using a hybrid approach of multi-heads attention-based control flow traces and image visualization

Android is the most widely used mobile platform, making it a prime target for malicious attacks. Therefore, it is imperative to effectively circumvent these attacks. Recently, machine learning has been a promising solution for malware detection, which relies on distinguishing features. While machine learning-based malware scanners have a large number of features, adversaries can avoid detection by using feature-related expertise. Therefore, one of the main tasks of the Android security industry is to consistently propose cutting-edge features that can detect suspicious activity. This study presents a novel feature representation approach for malware detection that combines API-Call Graphs (ACGs) with byte-level image representation. First, the reverse engineering procedure is used to obtain the Java programming codes and Dalvik Executable (DEX) file from Android Package Kit (APK). Second, to depict Android apps with high-level features, we develop ACGs by mining API-Calls and API sequences from Control Flow Graph (CFG). The ACGs can act as a digital fingerprint of the actions taken by Android apps. Next, the multi-head attention-based transfer learning method is used to extract trained features vector from ACGs. Third, the DEX file is converted to a malware image, and the texture features are extracted and highlighted using a combination of FAST (Features from Accelerated Segment Test) and BRIEF (Binary Robust Independent Elementary Features). Finally, the ACGs and texture features are combined for effective malware detection and classification. The proposed method uses a customized dataset prepared from the CIC-InvesAndMal2019 dataset and outperforms state-of-the-art methods with 99.27% accuracy.

Automatic Prediction of T2/T3 Staging of Rectal Cancer Based on Radiomics and Machine Learning

The staging of rectal cancer is very important to determine the treatment plans. This study investigated the relationship between the imaging features and the rectal cancer staging, so that the staging of rectal cancer can be automatically predicted based on the imaging features. A total of 81 patients who underwent with T2 or T3 stage rectal cancer from April 2018 to March 2019 were included. Firstly, tumor was labeled by the radiologist to outline the ROI (region of interest) in the high-resolution MRI images. Then the ROI was segmented by FCNN model and MedicalNet model. Secondly, features of the ROI were extracted by radiomics method. Thirdly, the key features were screened out from large number of features. Finally, a machine learning model was trained to predict rectal cancer stage. Two machine learning tools, back-projected neural network (BPNN) and support vector machine method (SVM) were used for the T2/T3 staging prediction of rectal cancer. The accuracy of our methods was 88.2%∼90.5% in the testing dataset, with a confidence interval of 95%, the sensitivity was 90.8%∼91.2%, the specificity was 85.9%∼87.6%, which were better than the traditional method. The area under the curve (AUC) of the BPNN method was 0.81 ± 0.01, which had better prediction performance than the SVM method (AUC = 0.75 ± 0.03). Some of the radiomics features have a significant relationship with the T2/T3 stage of rectal cancer, so it is possible to effectively predict the T2/T3 stage of rectal cancer using the selected radiomics features and machine learning methods.

Detection of Fall Risk in Multiple Sclerosis by Gait Analysis-An Innovative Approach Using Feature Selection Ensemble and Machine Learning Algorithms.

One of the common causes of falls in people with Multiple Sclerosis (pwMS) is walking impairment. Therefore, assessment of gait is of importance in MS. Gait analysis and fall detection can take place in the clinical context using a wide variety of available methods. However, combining these methods while using machine learning algorithms for detecting falls has not been performed. Our objective was to determine the most relevant method for determining fall risk by analyzing eleven different gait data sets with machine learning algorithms. In addition, we examined the most important features of fall detection. A new feature selection ensemble (FS-Ensemble) and four classification models (Gaussian Naive Bayes, Decision Tree, k-Nearest Neighbor, Support Vector Machine) were used. The FS-Ensemble consisted of four filter methods: Chi-square test, information gain, Minimum Redundancy Maximum Relevance and RelieF. Various thresholds (50%, 25% and 10%) and combination methods (Union, Union 2, Union 3 and Intersection) were examined. Patient-reported outcomes using specialized walking questionnaires such as the 12-item Multiple Sclerosis Walking Scale (MSWS-12) and the Early Mobility Impairment Questionnaire (EMIQ) achieved the best performances with an F1 score of 0.54 for detecting falls. A combination of selected features of MSWS-12 and EMIQ, including the estimation of walking, running and stair climbing ability, the subjective effort as well as necessary concentration and walking fluency during walking, the frequency of stumbling and the indication of avoidance of social activity achieved the best recall of 75%. The Gaussian Naive Bayes was the best classification model for detecting falls with almost all data sets. FS-Ensemble improved the classification models and is an appropriate technique for reducing data sets with a large number of features. Future research on other risk factors, such as fear of falling, could provide further insights.

Sparse Data Reconstruction, Missing Value and Multiple Imputation through Matrix Factorization

Social science approaches to missing values predict avoided, unrequested, or lost information from dense data sets, typically surveys. The authors propose a matrix factorization approach to missing data imputation that (1) identifies underlying factors to model similarities across respondents and responses and (2) regularizes across factors to reduce their overinfluence for optimal data reconstruction. This approach may enable social scientists to draw new conclusions from sparse data sets with a large number of features, for example, historical or archival sources, online surveys with high attrition rates, or data sets created from Web scraping, which confound traditional imputation techniques. The authors introduce matrix factorization techniques and detail their probabilistic interpretation, and they demonstrate these techniques’ consistency with Rubin’s multiple imputation framework. The authors show via simulations using artificial data and data from real-world subsets of the General Social Survey and National Longitudinal Study of Youth cases for which matrix factorization techniques may be preferred. These findings recommend the use of matrix factorization for data reconstruction in several settings, particularly when data are Boolean and categorical and when large proportions of the data are missing.

Distributed Fuzzy Cognitive Maps for Feature Selection in Big Data Classification

The features of a dataset play an important role in the construction of a machine learning model. Because big datasets often have a large number of features, they may contain features that are less relevant to the machine learning task, which makes the process more time-consuming and complex. In order to facilitate learning, it is always recommended to remove the less significant features. The process of eliminating the irrelevant features and finding an optimal feature set involves comprehensively searching the dataset and considering every subset in the data. In this research, we present a distributed fuzzy cognitive map based learning-based wrapper method for feature selection that is able to extract those features from a dataset that play the most significant role in decision making. Fuzzy cognitive maps (FCMs) represent a hybrid computing technique combining elements of both fuzzy logic and cognitive maps. Using Spark’s resilient distributed datasets (RDDs), the proposed model can work effectively in a distributed manner for quick, in-memory processing along with effective iterative computations. According to the experimental results, when the proposed model is applied to a classification task, the features selected by the model help to expedite the classification process. The selection of relevant features using the proposed algorithm is on par with existing feature selection algorithms. In conjunction with a random forest classifier, the proposed model produced an average accuracy above 90%, as opposed to 85.6% accuracy when no feature selection strategy was adopted.

A Modified Firefly Deep Ensemble for Microarray Data Classification

Abstract Many researchers are using microarray technology to examine and investigate the levels of gene expression in a specific organism, which is an emerging trend in the field of genetic research. Microarray studies have a wide range of applications in the health sector, including disease prediction and diagnostics, as well as cancer research. Due to the existence of irrelevant or duplicated data in microarray datasets, it is difficult to correctly and immediately capture possible patterns using existing algorithms. Feature selection (FS) has evolved into a critical approach for identifying and eliminating the most pertinent qualities. The enormous dimensionality of microarray datasets, on the other hand, presents a significant barrier to the majority of available FS techniques. In this research, we propose a Modified Firefly Feature Selection (MFFS) algorithm that will reduce the irrelevant attributes needed for classification and a Deep Learning Model for classifying the microarray data. The experimental outcomes show that the proposed MFFS algorithm combined with a Hybrid Deep Learning Algorithm outperforms the existing methods in terms of feature set size, accuracy, precision, recall, F-measure and AUC for a dataset with larger number of features.

Used Car Price Prediction Based on the Iterative Framework of XGBoost+LightGBM

To better address the problem of the low prediction accuracy of used car prices under a large number of features and big data and improve the accuracy of existing deep learning models, an iterative framework combining XGBoost and LightGBM is proposed in this paper. First, the relevant data processing is carried out for the initial recognition features. Then, by training the deep residual network, the predicted results are fused with the original features as new features. Finally, the new feature group is input into the iteration framework for training, the iteration is stopped, and the results are output when the performance reaches the highest value. These experimental results show that the combination of the deep residual network and iterative framework has a better prediction accuracy than the random forest and deep residual network. At the same time, by combining the existing mainstream methods with the iterative framework, it is verified that the iterative framework proposed in this paper can be applied to other models and greatly improve the prediction performance of other models.

SCNIC: Sparse correlation network investigation for compositional data.

Microbiome studies are often limited by a lack of statistical power due to small sample sizes and a large number of features. This problem is exacerbated in correlative studies of multi-omic datasets. Statistical power can be increased by finding and summarizing modules of correlated observations, which is one dimensionality reduction method. Additionally, modules provide biological insight as correlated groups of microbes can have relationships among themselves. To address these challenges, we developed SCNIC: Sparse Cooccurrence Network Investigation for compositional data. SCNIC is open-source software that can generate correlation networks and detect and summarize modules of highly correlated features. Modules can be formed using either the Louvain Modularity Maximization (LMM) algorithm or a Shared Minimum Distance algorithm (SMD) that we newly describe here and relate to LMM using simulated data. We applied SCNIC to two published datasets and we achieved increased statistical power and identified microbes that not only differed across groups, but also correlated strongly with each other, suggesting shared environmental drivers or cooperative relationships among them. SCNIC provides an easy way to generate correlation networks, identify modules of correlated features and summarize them for downstream statistical analysis. Although SCNIC was designed considering properties of microbiome data, such as compositionality and sparsity, it can be applied to a variety of data types including metabolomics data and used to integrate multiple data types. SCNIC allows for the identification of functional microbial relationships at scale while increasing statistical power through feature reduction.

Research on imbalance machine learning methods for MRT_1WI soft tissue sarcoma data

BackgroundSoft tissue sarcoma is a rare and highly heterogeneous tumor in clinical practice. Pathological grading of the soft tissue sarcoma is a key factor in patient prognosis and treatment planning while the clinical data of soft tissue sarcoma are imbalanced. In this paper, we propose an effective solution to find the optimal imbalance machine learning model for predicting the classification of soft tissue sarcoma data.MethodsIn this paper, a large number of features are first obtained based on T_1WI images using the radiomics methods.Then, we explore the methods of feature selection, sampling and classification, get 17 imbalance machine learning models based on the above features and performed extensive experiments to classify imbalanced soft tissue sarcoma data. Meanwhile, we used another dataset splitting method as well, which could improve the classification performance and verify the validity of the models.ResultsThe experimental results show that the combination of extremely randomized trees (ERT) classification algorithm using SMOTETomek and the recursive feature elimination technique (RFE) performs best compared to other methods. The accuracy of RFE+STT+ERT is 81.57% , which is close to the accuracy of biopsy, and the accuracy is 95.69% when using another dataset splitting method.ConclusionPreoperative predicting pathological grade of soft tissue sarcoma in an accurate and noninvasive manner is essential. Our proposed machine learning method (RFE+STT+ERT) can make a positive contribution to solving the imbalanced data classification problem, which can favorably support the development of personalized treatment plans for soft tissue sarcoma patients.

A novel SSD fault detection method using GRU-based Sparse Auto-Encoder for dimensionality reduction

In recent years, with the development of flash memory technology, storage systems in large data centers are typically built upon thousands or even millions of solid-state drives (SSDs). Therefore, the failure of SSDs is inevitable. An SSD failure may cause unrecoverable data loss or unavailable system service, resulting in catastrophic results. Active fault detection technologies are able to detect device problems in advance, so it is gaining popularity. Recent trends have turned toward applying AI algorithms based on SSD SMART data for fault detection. However, SMART data of new SSDs contains a large number of features, and the high dimension of data features results in poor accuracy of AI algorithms for fault detection. To tackle the above problems, we improve the structure of traditional Auto-Encoder (AE) based on GRU and propose an SSD fault detection method – GAL based on dimensionality reduction with Gated Recurrent Unit (GRU) sparse autoencoder (GRUAE) by combining the temporal characteristics of SSD SMART data. The proposed method trains the GRUAE model with SSD SMART data firstly, and then adopts the encoder of GRUAE model as the dimensionality reduction tool to reduce the original high-dimensional SSD SMART data, aiming at reducing the influence of noise features in original SSD SAMRT data and highlight the features more relevant to data characteristics to improve the accuracy of fault detection. Finally, LSTM is adopted for fault detection with low-dimensional SSD SMART data. Experimental results on real SSD dataset from Alibaba show that the fault detection accuracy of various AI algorithms can be improved by varying degrees after dimensionality reduction with the proposed method, and GAL performs best among all methods.

Feature Extraction and Recognition of Human Physiological Signals Based on the Convolutional Neural Network

Human physiological signal processing is one of the research fields widely used in recent years. Research on human physiological signals plays a vital role in predicting human health and detecting and classifying certain disease outbreaks. The network of human physiological signals is difficult to determine because it contains a lot of information about human activities. To this end, a variety of feature extraction, feature selection, and classification algorithms have been implemented in the anomaly prediction process. However, it has the main disadvantage of classification results, using a large number of features and increasing complexity. In order to solve these problems, this paper proposes a convolutional neural network-based extraction technique for human physiological signal features and uses an MPL classifier to detect whether the ECG signal is normal or not, taking the ECG signal as an example. In this paper, the signal preprocessing method based on wavelet transform and morphological filtering is adopted, and the high-frequency signal is removed by wavelet transform, and the low-frequency signal is removed by morphological filtering. A wide range of tests on ECG signals obtained from the MIT-BIH-AR databank and INCART database showed that the method has good detection performance with sensitivity Sen = 99.54%, positive prediction rate PPR = 99.65%, detecting mistake ratio DER = 0.35% and precision Acc = 99.55%, which is an improved performance compared to other techniques, proving the superiority of the present technique.

A novel Ontology-guided Attribute Partitioning ensemble learning model for early prediction of cognitive deficits using quantitative Structural MRI in very preterm infants

Structural magnetic resonance imaging studies have shown that brain anatomical abnormalities are associated with cognitive deficits in preterm infants. Brain maturation and geometric features can be used with machine learning models for predicting later neurodevelopmental deficits. However, traditional machine learning models would suffer from a large feature-to-instance ratio (i.e., a large number of features but a small number of instances/samples). Ensemble learning is a paradigm that strategically generates and integrates a library of machine learning classifiers and has been successfully used on a wide variety of predictive modeling problems to boost model performance. Attribute (i.e., feature) bagging method is the most commonly used feature partitioning scheme, which randomly and repeatedly draws feature subsets from the entire feature set. Although attribute bagging method can effectively reduce feature dimensionality to handle the large feature-to-instance ratio, it lacks consideration of domain knowledge and latent relationship among features. In this study, we proposed a novel Ontology-guided Attribute Partitioning (OAP) method to better draw feature subsets by considering the domain-specific relationship among features. With the better-partitioned feature subsets, we developed an ensemble learning framework, which is referred to as OAP-Ensemble Learning (OAP-EL). We applied the OAP-EL to predict cognitive deficits at 2 years of age using quantitative brain maturation and geometric features obtained at term equivalent age in very preterm infants. We demonstrated that the proposed OAP-EL approach significantly outperformed the peer ensemble learning and traditional machine learning approaches.