ReliefF Algorithm Research Articles

Advanced Diagnostic Technique for Alzheimer's Disease using MRI Top-Ranked Volume and Surface-based Features.

Alzheimer's disease (AD) is the most dominant type of dementia that has not been treated completely yet. Few Alzheimer's patients are correctly diagnosed on time. Therefore, diagnostic tools are needed for better and more efficient diagnoses. This study aimed to develop an efficient automated method to differentiate Alzheimer's patients from normal elderly and present the essential features with accurate Alzheimer's diagnosis. In this analytical study, 154 Magnetic Resonance Imaging (MRI) scans were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database, preprocessed, and normalized by the head size for extracting features (volume, cortical thickness, Sulci depth, and Gyrification Index Features (GIF). Relief-F algorithm, t-test, and one way-ANOVA were used for feature ranking to obtain the most effective features representing the AD for the classification process. Finally, in the classification step, four classifiers were used with 10 folds cross-validation as follows: Gaussian Support Vector Machine (GSVM), Linear Support Vector Machine (LSVM), Weighted K-Nearest Neighbors (W-KNN), and Decision Tree algorithm. The LSVM classifier and W-KNN produce a testing accuracy of 100% with only seven features. Additionally, GSVM and decision tree produce a testing accuracy of 97.83% and 93.48%, respectively. The proposed system represents an automatic and highly accurate AD detection with a few reliable and effective features and minimum time.

Early Monitoring of Cotton Verticillium Wilt by Leaf Multiple “Symptom” Characteristics

Early diagnosis of cotton verticillium wilt (VW) and accurate assessment of the disease degree are important prerequisites for preventing the large-scale development of cotton VW. Hyperspectral techniques have been widely used for monitoring the extent of plant diseases, but early detection of VW disease in cotton remains a challenge. In this study, the Boruta algorithm was used to select the key physiological characteristics (leaf temperature, chlorophyll a content, and equivalent water thickness) of cotton leaves at the early stage of VW disease, and then the Relief-F algorithm was used to select the spectral features indicating multiple “symptoms” of cotton VW disease at the early stage. In addition, a new cotton VW early monitoring indicator (CVWEI) was constructed by combining the weights of the new index and related bands using a hierarchical analysis (AHP) and entropy weighting method (EWM). The study showed that the physiological indices constructed under VW stress were better indicators of VW disease than traditional vegetation indices; CVEWI achieved a high accuracy of 95% in the test set, with a Kappa coefficient of 0.89; and the test set R2 was 0.73 and RMSE was 3.15% for monitoring disease severity, compared to the optimal classification constructed using a single spectral index. The results may provide new ideas and methods for early and accurate monitoring of VW and other fungal diseases.

A New Feature Analysis Approach to Selecting Channels of EEG for Fatigue Driving.

Fatigued driving is a significant contributor to traffic accidents. There are some issues with common EEG data of 32 channels, 64 channels, and 128 channels, such as difficult acquisition, high data redundancy, and difficult practical application. A new channel selection method called ReliefF_SFS is proposed to address the problem of how to reduce the number of channels while maintaining classification accuracy. It combines the ReliefF algorithm and the sequential forward selection (SFS) algorithm. When only T6, O1, Oz, T4, P3, and FC3 are used, the classification accuracy under Theta_Std+FE combined with ReliefF_SFS achieves 99.45%. The strategy suggested in this paper not only ensures the recognition accuracy but also reduces the number of channels when compared to other models based on the same data set.

Monopolar Grounding Fault Location Method of DC Distribution Network Based on Improved ReliefF and Weighted Random Forest

Compared with the pole-to-pole short circuit, the fault characteristics are not obvious when a monopolar grounding fault occurs in a DC distribution network, and it is difficult to locate the fault accurately. To solve this problem, this paper proposes a fault location method based on improved ReliefF and Weighted random forest (WRF). The 24 time and frequency-domain fault features of the postfault aerial mode current are calculated, and the most useful features are selected to form the optimal feature subset for input to the fault location estimator. In this paper, the ReliefF algorithm is utilized for automatic feature selection and obtaining the weights of features. In addition, the WRF algorithm is used to build the fault location estimator. Considering the fault location, fault resistance, noise and time window length, the Matlab/Simulink simulation platform is used to simulate the fault situation and compare it with other algorithms. The simulation results show that the average positioning error of the fault location method is less than 0.1%, which is not affected by the fault resistance and has strong robustness.

Identification of soybean based on Sentinel-1/2 SAR and MSI imagery under a complex planting structure

Soybeans, as the major global oil seed crops, are among the large-scale agricultural products imported into China. In this way, the accurate identification of their growing areas is the basis for agricultural decision-making and planting structure adjustment, which are of great significance to the national food security. Currently, there is insufficient research on the identification of soybean, using remote sensing, under complex agricultural planting structures. In this paper, Guoyang County, a typical soybean-producing region in the north of Anhui Province, China, was selected as the study area, and then the multi-temporal Sentinel-1/2 (S-1/2) microwave and optical multispectral data were integrated to obtain the spatial distribution of the planting areas, via a stepwise hierarchical extraction strategy. Considering the Jeffries-Matusita (JM) distance, August 18 (i.e., the early pod setting stage) was correspondingly determined as more appropriate for soybean extraction, compared with other time phases. For this purpose, a set of rules were established to eliminate the non-cropland pixels, and the total distribution of the field vegetation was derived accordingly. In total, 30 candidate features (viz. 10 spectral bands, 11 vegetation indices, 4 texture features, and 5 microwave polarization features) were selected, and their importance was evaluated based on the ReliefF algorithm. With the support of the typical cover type samples, the weight assessment method for the ReliefF feature was further bundled with three machine-learning (ML) algorithms, namely, the random forest (RF), the back-propagation neural network (BPNN), and the support vector machine (SVM), to single out the optimum subset of the features for soybean identification. The results finally revealed that the ReliefF-RF model outperformed the other two. With nine optimum features, Kappa coefficient also reached up to 0.77–0.85. Furthermore, the extraction accuracy produced by the optimum subset of the features was significantly greater than the original 10 spectral bands (Kappa coefficient of 0.75–0.84), albeit slightly lower than the effect of the total 30 features (Kappa coefficient of 0.78–0.85).

Study and Application of an Elevator Failure Monitoring System Based on the Internet of Things Technology

To prevent the occurrence of elevator safety accidents, in this study, an Internet of things-based elevator failure monitoring system is investigated. First, it introduces the Internet of things technology, preprocesses the relevant data, and extracts the features of the elevator operation data continuously collected by the elevator sensors. The objective of this paper is to study the application of IoT in elevators. Method. The Relief-F algorithm is used to evaluate the potential influencing factors. The results show that, as the batch size increases, the accuracy rate will gradually increase, but after more than 20, the accuracy rate will decrease. When the batch size is 20, the training result is the best. It can be seen that, as the time step increases, the accuracy of the prediction will be significantly improved. When the time step is 24, the prediction accuracy is the highest; after 24, the prediction rate will decline. In the diagram of the influence of learning rate on the model, the blue line indicates that the learning rate is 1.0, the red line is 0.1, and the black line is 0.01. With the increase of the iteration times, the effect is the best when the learning rate is 0.1. This system has high research value and broad application prospects and can make up for the current lack of monitoring in the elevator industry.

Shape selection recognition and scattering distribution prediction of adhesion targets in multi-scale dual-energy X-ray images of coal and gangue

ABSTRACT Identifying and predicting the distribution of scattered coal and gangue is the premise of locating them. By analyzing the shape of multi-scale coal and gangue, an approach for shape selection and recognition is provided. The issues of independent and adhesion target recognition, adhesion type recognition, and distribution prediction have been resolved. The binary target, circumscribed convex hull, and concave defect images are used to extract a total of 27 shape features. The ReliefF algorithm is employed to select features. The shape recognition model 1 shows the highest adhesion and independent recognition rates of 98.54%. For adhesion kinds, shape recognition model 2 gets the highest recognition accuracy of 92%. According to the experimental findings, the difference between predicted and actual results for the parameters describing the dispersed distribution of targets, such as target number, distribution overlap rate, and distribution density, is less than 4.1%, which is acceptable for use in real-world scenarios.

A new phenology-based method for mapping wheat and barley using time-series of Sentinel-2 images

In recent years, various techniques have been developed to generate crop-type maps based on remote sensing data. Wheat and barley are two major cereal crops cultivated as the first and fourth largest grain crops across the globe. The variations in spectral temporal profile of both crops are generally insignificant at small scales and therefore the two crops are phenologically fairly clearly separated; however, at large scale areas the variance of phenological parameters increases for both crops due to the effects of various climatic and orographic factors which adversely influences discrimination of wheat and barley. Additionally, wheat and barley are usually cultivated as both spring and winter or early and late season crops in some areas, making it more difficult to distinguish them. Therefore, developing a new method based on remote sensing data for effective discrimination of wheat and barley is an important necessity in the field of precision agriculture. To this end, this research presents a new phenology-based method to discriminate barley from wheat. In this study, Sentinel-2 (S2) time-series data of a study site in Iran (Markazi) and two sites in the USA (Idaho and North California), are employed. Spectral reflectance values of wheat and barley are examined during the growing season and a new spectral-temporal feature is successfully developed for automatic identification of the barley heading date. The Relief-f algorithm is then employed to select appropriate spectral features of S2 to distinguish wheat from barley at the heading date. Finally, generated spectral features at the heading date are used as input to Support Vector Machine (SVM) and Random Forest (RF) to produce barley and wheat maps. The Kappa coefficient and overall accuracy (OA) obtained for the three study sites are more than 0.67 and 76%, respectively. The findings of this study demonstrate the potential of remote sensing data to identify the phenological growth stages of barley and distinguish it successfully from wheat.

Investigation on the Identity Construction of Young Foreign Language Teachers in Colleges and Universities Based on Feature Selection Algorithm

With the gradual implementation of the strategic goal of strengthening the country in education, people’s attention to students’ learning is also increasing day by day. At the same time, the teaching effect of individual teachers depends to a large extent on the identity of teachers themselves, so it is necessary to improve the construction of teachers’ identity. Most of the research on teachers’ identity construction starts from the theory of identity, which does not play a significant role in the critical application in practical teaching activities. In response to these problems, this paper will use the ReliefF algorithm and the Spark algorithm in the feature selection algorithm to scientifically process the identity construction, and implement the application of the improved ReliefF feature selection algorithm and the feature selection application steps based on the Spark algorithm respectively. The experimental results show that the improved ReliefF algorithm has better feature selection accuracy when the feature range is 30% to 40%. It shows that the construction of teacher identity based on feature selection algorithm can provide an objective basis for the realization of teacher identity.

A novel random multi-subspace based ReliefF for feature selection

Feature selection is an important preprocessing technology for dimensionality reduction, which reduces the dimension of the dataset by acquiring a subset of features with the largest amount of information, and improves the classification accuracy to the greatest extent at the same time. Although different types of feature selection algorithms have achieved remarkable success, most of them lack the ability to mine information in different subspaces, and ignore the useful information contained in the abundant samples. In this research, a novel random multi-subspace based ReliefF (RBEFF) is proposed for feature selection. In this method, firstly, multiple feature partitions containing a large number of random subspaces with the same size are generated. Secondly, the ReliefF algorithm is used in each random subspace to obtain the local weight of the feature. The local weight vectors of random subspaces in each feature partition are combined to obtain the full weight vector. Finally, the full weight vectors of multiple feature partitions are integrated into the final weight vector, which contains the final weight of each feature in the original feature space feature. The feature selection is carried out dynamically according to the final weight vector. We evaluated the performance of the RBEFF on 28 UCI datasets with different sizes and compare RBEFF with 6 feature selection algorithms using KNN and DT classifiers’ three evaluation indicators. The comparisons and experimental results demonstrate the effectiveness, competitiveness, and superiority of RBEFF in solving feature selection problems.

Research on Common Fault Diagnosis and Classification Method of Centrifugal Pump Based on ReliefF and SVM

The centrifugal pump is an important rotating machine and it is very critical to identify and differentiate among its common faults as quickly and accurately as possible. Based on the ReliefF algorithm and the sparrow search algorithm (SSA) in conjunction with support vector machine (SVM), an approach for faults classification and diagnosis of centrifugal pumps is proposed, its advantages over traditional fault diagnosis methods include a reduction in the number of characteristic parameters, shorter diagnosis times, as well as improved classification accuracy and robustness. We collected the fault data by designing a centrifugal pump fault test bench that recorded vibration signals for the rotor misalignment fault, the rotor unbalance fault, the seal ring wear fault, and normal operating conditions, and preprocessed the collected signals with Kalman filtering to remove noise interference, the time domain characteristic indexes and the frequency domain characteristic indexes of the filtered signal were extracted, each feature index is given a distinct weight using the ReliefF method, and the eigenvalues with weights less than the threshold are deleted, and the feature indexes that remain create a defect feature matrix. Particle swarm optimization (PSO), genetic algorithm (GA), and simulated annealing algorithm (SA) were used to optimize the SVM for comparison in order to verify the SSA-SVM model’s performance for fault diagnosis. The comparison results show that the model has high recognition accuracy, short Classification time, and strong robustness.

Predicting COVID-19 Infection Using Machine Learning Methods Combined with Feature Selection

COVID-19 is an infection that has affected the world since December 31, 2019, and was declared a pandemic by WHO in March 2020. In this study, Multi-Layer Perceptron (MLP), Tree Boost (TB), Radial Basis Function Network (RBF), Support Vector Machine (SVM), and K-Means Clustering (kMC) individually combined with minimum redundancy maximum relevance (mRMR) and Relief-F have been used to construct new feature selection-based COVID-19 prediction models and discern the influential variables for prediction of COVID-19 infection. The dataset has information related to 20.000 patients (i.e., 10.000 positives, 10.000 negatives) and includes several personal, symptomatic, and non-symptomatic variables. The accuracy, recall, and F1-score metrics have been used to assess the models’ performance, whereas the generalization errors of the models were evaluated using 10-fold cross-validation. The results show that the average performance of mRMR is slightly better than Relief-F in predicting the COVID-19 infection of a patient. In addition, mRMR is more successful than the Relief-F algorithm in finding the relative relevance order of the COVID-19 predictors. The mRMR algorithm emphasizes symptomatic variables such as fever and cough, whereas the Relief-F algorithm highlights non-symptomatic variables such as age and race. It has also been observed that, in general, MLP outperforms all other classifiers for predicting the COVID-19 infection.

Incipient Fault Detection of Helical Gearbox Based on Variational Mode Decomposition and Time Synchronous Averaging

As the fault detection methods diagnose defects in the earlier stage, the subsequent costs will be reduced. Feature extraction from the vibration signal is the foremost step for incipient fault detection of gearboxes. However, the current statistical features in the time and frequency domains cannot diagnose the early or low intensity faults. In this research, for the first time, besides these features, some other features are extracted by combining variational mode decomposition and time synchronous average (VMD-TSA) to overcome the problem. The combinations have occurred in two ways. First, the Intrinsic Mode Functions (IMFs) of the TSA signal are calculated by VMD, and the Amplitude Energy (AE) and Permutation Entropy (PE) of the first four IMFs are computed. Secondly, the IMFs of vibration signals are calculated, and the TSA features are extracted from the most informative IMF. Moreover, 16 features in the time domain, 13 features in the frequency domain, and 9 features by TSA are extracted from the vibration signals. These features are extracted from healthy and four faulty conditions: crack, spalling, chipping, and wear in three different severities. After feature extraction, the Relief-F algorithm selects the informative features, and selected features are utilized for fault detection by a Feed-Forward Neural network (FNN) classifier. In this study, the ability of the VMD-TSA method is compared with others like Empirical Mode Decomposition-TSA (EMD-TSA) and Ensemble Empirical Mode Decomposition-TSA (EEMD-TSA), which shows that the proposed method is more powerful than others in early fault detection. Besides, the classification accuracy of these methods is compared with some other feature selection methods like Laplacian Score (LS), Principal Component Analysis (PCA), and Minimum Redundancy-Maximum Relevance (MRMR). Also, the performance of the FNN classifier is compared with the Support Vector Machine (SVM). As shown in this study, the VMD-TSA features improve the early fault detection in a positive manner. For instance, the classification accuracy of all features without VMD-TSA in crack fault detection is 93.98%. However, by adding VMD-TSA features, the accuracy grows up to 99.48% in the same conditions.

Classification of Glaucoma Based on Elephant-Herding Optimization Algorithm and Deep Belief Network

This study proposes a novel glaucoma identification system from fundus images through the deep belief network (DBN) optimized by the elephant-herding optimization (EHO) algorithm. Initially, the input image undergoes the preprocessing steps of noise removal and enhancement processes, followed by optical disc (OD) and optical cup (OC) segmentation and extraction of structural, intensity, and textural features. Most discriminative features are then selected using the ReliefF algorithm and passed to the DBN for classification into glaucomatous or normal. To enhance the classification rate of the DBN, the DBN parameters are fine-tuned by the EHO algorithm. The model has experimented on public and private datasets with 7280 images, which attained a maximum classification rate of 99.4%, 100% specificity, and 99.89% sensitivity. The 10-fold cross validation reduced the misclassification and attained 98.5% accuracy. Investigations proved the efficacy of the proposed method in avoiding bias, dataset variability, and reducing false positives compared to similar works of glaucoma classification. The proposed system can be tested on diverse datasets, aiding in the improved glaucoma diagnosis.

Semi-supervised Learning Based EEG Detection Approach for Rehabilitation Engineering

<p>A semi-supervised learning based EEG signal detection method was studied in this paper. The feature engineering system of this paper was established, which contains novel AutoEncoders mapping features. The optimal channel combination for all subjects was determined to improve recognition accuracy by ReliefF algorithm and recursive feature elimination. What’s more, the semi-supervised learning method based on pseudo-labelling was introduced to the character recognition method, in which the training samples were dynamically reorganized and updated, so that the proposed method could complete the symbol recognition with limited number of training samples. Based on the features extacted and the optimal channel combination, the recognition accuracy of the character recognition method can reach up to 100%.</p> <p> </p>

Machine Learning-Based Vehicle Model Construction and Validation—Toward Optimal Control Strategy Development for Plug-In Hybrid Electric Vehicles

Advances in machine learning inspire novel solutions for the validation of complex vehicle models and spur an easy manner to promote energy management performance of complexly configured vehicles, such as plug-in hybrid electric vehicles (PHEVs). A constructed PHEV model, based on the four-wheel-drive passenger vehicle configuration, is validated through an efficient virtual test controller (VTC) developed in this article. The VTC is designed via a novel approach based on the least-square support vector machine and random forest with the inner-interim data filtered by the ReliefF algorithm to validate the vehicle model as necessary. This article discusses the process and highlights the accuracy improvements of the PHEV model that is achieved by implementing the VTC. The validity of the VTC is addressed by examining the PHEV model to mimic the characteristics of internal combustion engine, motor, and generator behaviors observed through the benchmark test. Sufficient simulations and hardware-in-loop test are employed to demonstrate the capability of the novel VTC-based model validation method in practical applications. The major novelty of this article lies in the development of a VTC, by which the vehicle model can be efficiently developed, providing solid framework and enormous convenience for control strategy design.

An Optimal Control Strategy for Plug-In Hybrid Electric Vehicles Based on Enhanced Model Predictive Control With Efficient Numerical Method

Advances in machine learning inspire novel solutions for the validation of complex vehicle models and spur an easy manner to promote energy management performance of complexly configured vehicles, such as plug-in hybrid electric vehicles (PHEVs). A constructed PHEV model, based on the four-wheel-drive passenger vehicle configuration, is validated through an efficient virtual test controller (VTC) developed in this article. The VTC is designed via a novel approach based on the least-squares support vector machine and random forest with the inner-interim data filtered by the ReliefF algorithm to validate the vehicle model as necessary. This article discusses the process and highlights the accuracy improvements of the PHEV model that is achieved by implementing the VTC. The validity of the VTC is addressed by examining the PHEV model to mimic the characteristics of the internal combustion engine, motor, and generator behaviors observed through the benchmark test. Sufficient simulations and hardware-in-the-loop tests are employed to demonstrate the capability of the novel VTC-based model validation method in practical applications. The major novelty of this article lies in the development of a VTC, by which the vehicle model can be efficiently developed, providing a solid framework and enormous convenience for control strategy design.

A Multi-Scale Residential Areas Matching Method Considering Spatial Neighborhood Features

Residential areas is one of the basic geographical elements on the map and an important content of the map representation. Multi-scale residential areas matching refers to the process of identifying and associating entities with the same name in different data sources, which can be widely used in map compilation, data fusion, change detection and update. A matching method considering spatial neighborhood features is proposed to solve the complex matching problem of multi-scale residential areas. The method uses Delaunay triangulation to divide complex matching entities in different scales into closed domains through spatial neighborhood clusters, which can obtain many-to-many matching candidate feature sets. At the same time, the geometric features and topological features of the residential areas are fully considered, and the Relief-F algorithm is used to determine the weight values of different similarity features. Then the similarity and spatial neighborhood similarity of the polygon residential areas are calculated, after which the final matching results are obtained. The experimental results show that the accuracy rate, recall rate and F value of the matching method are all above 90%, which has a high matching accuracy. It can identify a variety of matching relationships and overcome the influence of certain positional deviations on matching results. The proposed method can not only take account of the spatial neighborhood characteristics of residential areas, but also identify complex matching relationships in multi-scale residential areas accurately with a good matching effect.

Optimized adaptive neuro-fuzzy inference system based on hybrid grey wolf-bat algorithm for schizophrenia recognition from EEG signals.

Schizophrenia is a chronic mental disorder that impairs a person's thinking capacity, feelings and emotions, behavioural traits, etc., Emotional distortions, delusions, hallucinations, and incoherent speech are all some of the symptoms of schizophrenia, and cause disruption of routine activities. Computer-assisted diagnosis of schizophrenia is significantly needed to give its patients a higher quality of life. Hence, an improved adaptive neuro-fuzzy inference system based on the Hybrid Grey Wolf-Bat Algorithm for accurate prediction of schizophrenia from multi-channel EEG signals is presented in this study. The EEG signals are pre-processed using a Butterworth band pass filter and wICA initially, from which statistical, time-domain, frequency-domain, and spectral features are extracted. Discriminating features are selected using the ReliefF algorithm and are then forwarded to ANFIS for classification into either schizophrenic or normal. ANFIS is optimized by the Hybrid Grey Wolf-Bat Algorithm (HWBO) for better efficiency. The method is experimented on two separate EEG datasets-1 and 2, demonstrating an accuracy of 99.54% and 99.35%, respectively, with appreciable F1-score and MCC. Further experiments reveal the efficiency of the Hybrid Wolf-Bat algorithm in optimizing the ANFIS parameters when compared with traditional ANFIS model and other proven algorithms like genetic algorithm-ANFIS, particle optimization-ANFIS, crow search optimization algorithm-ANFIS and ant colony optimization algorithm-ANFIS, showing high R2 value and low RSME value. To provide a bias free classification, tenfold cross validation is performed which produced an accuracy of 97.8% and 98.5% on the two datasets respectively. Experimental outcomes demonstrate the superiority of the Hybrid Grey Wolf-Bat Algorithm over the similar techniques in predicting schizophrenia.

Prediction and molecular mechanism of phosphate adsorption by metal oxides

<p indent="0mm">Metal oxide adsorbents have emerged as promising phosphate removal materials due to their excellent adsorption performance, low price, and stability. However, the relationship between the structure of metal oxides and the performance of phosphate adsorption is still unclear, and the molecular mechanism of phosphate adsorption by metal oxides remains to be elucidated. Understanding the major influencing factors and mechanisms of phosphate adsorption by metal oxides is critical for the design of metal-based materials for phosphate adsorption. In this study, the kinetics and thermodynamics of phosphate adsorption by Gd₂O₃ and CeO₂ were analyzed, and the phosphate adsorption capacity of 32 metal oxides was determined. ReliefF algorithm was used to identify the key descriptors of metal oxides affecting phosphate adsorption. Based on the descriptors, prediction models for phosphate adsorption were developed based on six machine learning algorithms, and the applicability domain of the models was characterized. Phosphate adsorption on the surface of six typical metal oxides was investigated by using molecular simulations based on density functional theory (DFT). The results demonstrated that the maximum adsorption capacity of Gd₂O₃ for phosphate was 4.14 times higher than that of CeO₂. The crystal structure of Gd₂O₃ was changed after adsorbing phosphate and resulted in Gd₂O₃ transformation into urchin shaped structures, but the crystal structure of CeO₂ remained basically unchanged before and after adsorption. The results of phosphate adsorption of 32 metal oxides showed that rare earth metal oxides (except CeO₂) and ZnO had higher adsorption capacity than other metal oxides. The electronegativity and cationic charge of metal oxides were identified as two important factors for phosphate adsorption using ReliefF algorithm. Six machine learning algorithms were utilized to build the prediction models for screening metal oxides with high adsorption capacity, and the predictive accuracy of the decision tree model exceeded 90% for both training and test sets. Classification rules of the decision tree model indicated that metal oxides with electronegativity ≤ 1.23 had high adsorption potential for phosphate. Characterization of the model’s applicability domain showed that all metal oxides were within the applicability domain. DFT analysis showed that adsorption energies of Gd₂O₃ and Y₂O₃ for phosphate were −3.97 and <sc>−4.56 eV,</sc> which were higher than those of CeO₂, NiO, ZnO and Cr₂O₃<sc>(–1.61− –0.26 eV).</sc> The adsorption configuration of Gd₂O₃ and Y₂O₃ with higher adsorption energy for phosphate was bidentate binuclear, while that of CeO₂, NiO, ZnO and Cr₂O₃ with lower adsorption energy was monodentate mononuclear. Combined with molecular orbital theory and density of states analysis, it was found that the d orbitals from Gd and Y atoms higher than the Fermi level interact with the p orbitals of the oxygen atom from the phosphate, splitting into antibonding orbitals higher than the Fermi level. Theoretical calculations revealed a new mechanism that the d orbitals energy level of metal atoms determines the phosphate adsorption energy. The prediction model established in this study can provide a basis for screening metal-based materials that can effectively adsorb phosphate, and the molecular mechanism deciphered by DFT calculations can provide a theoretical basis for the design of adsorbents.