feature selection

Traffic accidents remain a critical global public health issue, resulting in numerous fatalities and injuries annually.

This study aims to explore the application of machine learning (ML) in analyzing traffic accident data obtained from self-report questionnaires to identify factors influencing the incidence and severity of accidents.

The study design is cross-sectional. In this study, approximately 660 participants completed the questionnaire, of which 43 were incomplete or invalid and were excluded. The remaining 617 participants answered all questions in full. Participants were selected using a convenience sampling method from five districts in Shiraz to ensure diversity, including outreach to taxi and heavy vehicle terminals. Data were collected through face-to-face questionnaires administered by trained researchers, and all responses were self-reported. The dataset collected from 617 participants includes information on demographics, vehicle and road features, personality traits, driving habits, and risky driving behavior. The questionnaire incorporated multiple validated instruments capturing driving behavior, demographics (such as age, gender, marital status, education, income), and habits (e.g., driving duration, cellphone use, fatigue, and substance use). Various ML algorithms, such as random forest and SHapley Additive exPlanations (SHAP) analysis, were employed to identify factors influencing both the occurrence and severity of accidents. Furthermore, the C5.0 algorithm was utilized to extract specific patterns, while prediction tasks were addressed using a combination of random forest, support vector machine (SVM), logistic regression, and Naive Bayes algorithms.

The random forest algorithm highlighted that factors such as income, driving time, working time, age, duration of non-stop driving, type of law enforcement, openness, normlessness, sensation seeking, and vehicle safety significantly influence the occurrence of accidents. For accident severity, important predictors included driving time, non-stop driving, working time, age, aggressive violations, income, road quality, type of law enforcement, driving while tired, vehicle safety, foreign car status, and vehicle comfort. Additionally, the C5.0 algorithm revealed specific patterns—such as high normlessness and extended driving hours—increasing the likelihood of accidents, while factors like low normlessness and balanced income served as protective elements.

The findings highlight the impact of lifestyle and work-related factors, as well as certain personality traits of drivers, on the incidence and severity of accidents. While the results of the study should not be taken verbatim due to the reliance on self-reported data, the study supports the application of ML in the analysis of accident data. It also advocates for the use of strategies including social and economic interventions, psychological assessments, enhanced road safety education, and customized regulatory measures based on individual risk assessments to effectively prevent traffic accidents.

High-dimensional datasets often contain an abundance of features, many of which are irrelevant to the subject of interest. This issue is compounded by the frequently low number of samples and imbalanced class samples. These factors can negatively impact the performance of classification algorithms, necessitating feature selection before classification. The primary objective of feature selection algorithms is to identify a minimal subset of features that enables accurate classification.

In this paper, we propose a two-stage hybrid method for the optimal selection of relevant features. In the first stage, a filter method is employed to assign weights to the features, facilitating the removal of redundant and irrelevant features and reducing the computational cost of classification algorithms. A subset of high-weight features is retained for further processing in the second stage. In this stage, an enhanced Harris Hawks Optimization algorithm and GRASP, augmented with crossover and mutation operators from genetic algorithms, are utilized based on the weights calculated in the first stage to identify the optimal feature set.

Experimental results demonstrate that the proposed algorithm successfully identifies the optimal subset of features.

The two-stage hybrid method effectively selects the optimal subset of features, improving the performance of classification algorithms on high-dimensional datasets. This approach addresses the challenges posed by the abundance of features, low number of samples, and imbalanced class samples, demonstrating its potential for application in various fields.

This study aimed to develop a hybrid model for variable selection in high-dimensional survival analysis using a support vector regression (SVR), to identify prognostic biomarkers associated with survival in oral cancer (OC) patients through the analysis of gene expression data. 

In this retrospective cohort study, gene expression profiles (54,613 probes) related to 97 patients from the GSE41613 dataset from the GEO repository were used. First of all, martingale residuals were obtained using a Cox regression without covariates, and were used as pseudo-survival outcome. Then, the particle swarm optimization (PSO) and genetic algorithm (GA) were used in combination with SVR for selecting features related to pseudo-survival outcome. Concordance index (C-index), mean absolute error (MAE), mean squared error (MSE) and R-squares, were used to evaluate the performance of the models using selected features. Functional enrichment analysis was performed using DAVID database, and external validation utilized three independent datasets (GSE9844, GSE75538, GSE37991, GSE42743). 

The findings indicated that the PSO-based method outperformed the GA-based method, achieving a smaller MAE (0.061) and MSE (0.005), R-square (0.99) and C-index (0.973), selecting 291 probes from 1069 screened. A protein-protein interaction (PPI) network was constructed, including 200 nodes and 120 edges. Eleven key genes with the highest degree, including RBM25, SMC3, PRPF40A, POLE, SRRT, BCLAF1, PDS5B, HNRNPR, JAK1, MED23, and SULT1A1 were identified as significant biomarkers associated with OC survival. 

The PSO-based hybrid model effectively improved SVR performance in survival prediction for OC patients and identified key prognostic biomarkers. Despite its promising results and validation on independent datasets, limitations in generalizability and signs of overfitting suggest the model is not yet ready for clinical use. Further studies with larger, diverse datasets are recommended.

Parkinson’s disease (PD) is a neurodegenerative disorder that progressively worsens with age, particularly affecting the elderly. Symptoms of PD include visual hallucinations, depression, autonomic dysfunction, and motor difficulties. Conventional diagnostic methods often rely on subjective interpretations of movement, which can be subtle and challenging to assess accurately, potentially leading to misdiagnoses. However, recent studies indicate that over 90% of individuals with PD exhibit vocal abnormalities at the onset of the disease. Machine learning (ML) techniques have shown promise in addressing these diagnostic challenges due to their higher efficiency and reduced error rates in analyzing complex, high-dimensional datasets, particularly those derived from speech signals. This study investigates 12 machine learning models—logistic regression (LR), support vector machine (SVM, linear/RBF), K-nearest neighbor (KNN), Naïve bayes (NB), decision tree (DT), random forest (RF), extra trees (ET), gradient boosting (GbBoost), extreme gradient boosting (XgBoost), adaboost, and multi-layer perceptron (MLP)—to develop a robust ML model capable of reliably identifying PD cases. The analysis utilized a PD voice dataset comprising 756 acoustic samples from 252 participants, including 188 individuals with PD and 64 healthy controls. The dataset included 130 male and 122 female subjects, with age ranges of 33 - 87 years and 41 - 82 years, respectively. To enhance model performance, the GridSearchCV method was employed for hyperparameter tuning, alongside recursive feature elimination (RFE) and minimum redundancy maximum relevance (mRMR) feature selection techniques. Among the 12 ML models evaluated, the RF model with the RFE-generated feature subset (RFE-50) emerged as the top performer. It achieved an accuracy of 96.46%, a recall of 0.96, a precision of 0.97, an F1-score of 0.96, and an AUC score of 0.998, marking the highest performance metrics reported for this dataset in recent studies.

This study explores a novel approach to detecting arousal levels through the analysis of electroencephalography (EEG) signals. Leveraging the Faller database with data from 18 healthy participants, we employ a 64‑channel EEG system.

The approach we employ entails the extraction of ten frequency characteristics from every channel, culminating in a feature vector of 640 dimensions for each signal instance. To enhance classification accuracy, we employ a genetic algorithm for feature selection, treating it as a multiobjective optimization task. The approach utilizes fast bit hopping for efficiency, overcoming traditional bit‑string limitations. A hybrid operator expedites algorithm convergence, and a solution selection strategy identifies the most suitable feature subset.

Experimental results demonstrate the method’s effectiveness in detecting arousal levels across diverse states, with improvements in accuracy, sensitivity, and specificity. In scenario one, the proposed method achieves an average accuracy, sensitivity, and specificity of 93.11%, 98.37%, and 99.14%, respectively. In scenario two, the averages stand at 81.35%, 88.65%, and 84.64%.

The obtained results indicate that the proposed method has a high capability of detecting arousal levels in different scenarios. In addition, the advantage of employing the proposed feature reduction method has been demonstrated.

The remarkable growth of lung cancer and its associated impacts and consequences, along with the substantial costs it imposes on society, has driven the medical community to pursue programs aimed at further examination, prevention, early detection, and diagnosis. In medicine science, timely discovery and diagnosis of diseases can prevent many life-threatening conditions and save people's lives.

This study aims to predict lung cancer using a novel feature selection method integrated with a classifier. Our approach entails a comprehensive four-stage method. Initially, we calculate feature similarities within a lung cancer dataset using the absolute value of the Pearson correlation coefficient, followed by the clustering of initial features using the community detection algorithm called Louvain. Next, we employ techniques to determine the optimal subset of features using the concept of node centrality. Ultimately, lung cancer diagnosis is executed using the selected features, leveraging a classifier.

Comparative analysis reveals that our proposed method outperforms existing techniques in terms of reduced execution time and improved prediction accuracy. When compared with established methods, our approach demonstrates superior outcomes in terms of the number of selected features and classification accuracy. Our method reduced 12600 features to 118 features and its accuracy was 95.28, 95.49, 95.23 and 95.32 for Support Vector Machine (SVM), Decision Tree (DT), Naive Bayes (NB) and K-Nearest Neighbor (KNN) classifier. The comparison of runtime shows that the proposed method is significantly improved with a runtime of 2.146 seconds compared to other methods.

The proposed feature selection method successfully reduced the initial feature set and significantly decreased computational time. Moreover, the achieved prediction accuracies underscore the reliability of our approach. This significant reduction in feature space while maintaining consistently high prediction accuracies serves as a strong validation of the potency and practical applicability of our methodology in the domain of lung cancer prediction. These compelling results strongly advocate for the potential real-world impact of our approach.

Applying efficient feature extraction and selection methods is essential in improving the performance of machine learning algorithms employed in brain-computer interface (BCI) systems.

The current study aims to enhance the performance of a motor imagery-based BCI by improving the feature extraction and selection stages of the machine-learning algorithm applied to classify the different imagined movements.

In this study, a multi-rate system for spectral decomposition of the signal is designed, and then the spatial and temporal features are extracted from each sub-band. To maximize the classification accuracy while simplifying the model and using the smallest set of features, the feature selection stage is treated as a multiobjective optimization problem, and the Pareto optimal solutions of these two conflicting objectives are obtained. For the feature selection stage, non-dominated sorting genetic algorithm II (NSGA-II), an evolutionary-based algorithm, is used wrapper-based, and its effect on the BCI performance is explored. The proposed method is
implemented on a public dataset known as BCI competition III dataset IVa.

Extracting the spatial and temporal features from different sub-bands and selecting the features with an evolutionary optimization approach in this study led to an improved classification accuracy of 92.19% which has a higher value compared to the state of the art.

The results show that the proposed improved classification accuracy could achieve a high-performance subject-specific BCI system.

Previous research has identified key factors affecting in vitro fertilization or intracytoplasmic sperm injection success, yet the lack of a standardized approach for various treatments remains a challenge.

The objective of this study is to utilize a machine learning approach to identify the principal predictors of success in in vitro fertilization and intracytoplasmic sperm injection treatments.

We collected data from 734 individuals at 2 infertility centers in Mashhad, Iran between November 2016 and March 2017. We employed feature selection methods to reduce dimensionality in a random forest model, guided by hesitant fuzzy sets (HFSs). A hybrid approach enhanced predictor identification and accuracy (ACC), as assessed using machine learning metrics such as matthew’s correlation coefficient, runtime, ACC, area under the receiver operating characteristic curve, precision or positive predictive value, recall, and F-Score, demonstrating the effectiveness of combining feature selection methods.

Our hybrid feature selection method excelled with the highest ACC (0.795), area under the receiver operating characteristic curve (0.72), and F-Score (0.8), while selecting only 7 features. These included follicle-stimulation hormone (FSH), 16Cells, FAge, oocytes, quality of transferred embryos (GIII), compact, and unsuccessful.

We introduced HFSs in our novel method to select influential features for predicting infertility success rates. Using a multi-center dataset, HFSs improved feature selection by reducing the number of features based on standard deviation among criteria. Results showed significant differences between pregnant and non-pregnant groups for selected features, including FSH, FAge, 16Cells, oocytes, GIII, and compact. We also found a significant correlation between FAge and fetal heart rate and clinical pregnancy rate, with the highest FSH level (31.87%) observed for doses ranging from 10-13 (mIU/ml).

It is important to diagnose coronavirus disease 2019 (COVID-19) caused by severe acute respiratory syndrome coronavirus 2 at an early stage and to monitor severely infected patients in order to reduce the lethality of the disease. In addition, there is a need for alternative methods with lower costs and faster results to determine the severity of the disease. In this context, routine blood values can be used to determine the diagnosis/prognosis and mortality of COVID-19. In this study, three optimized datasets were prepared to determine the features that affect the diagnosis, prognosis, and mortality of COVID-19. These datasets can be used by researchers to determine the diagnosis and severity of COVID-19 with various classifier machine learning models and artificial intelligence methods. It is hoped that studies on these datasets will reduce the negative pressures on the health system and provide important clinical guidance for decision-makers in the diagnosis and prognosis of COVID-19.

Age-related macular degeneration (AMD) is the progressive degenerative disease of the macula and the main cause of blindness in older adults. Various risk factors have been associated with disease progression among different individuals.AMD is affected by different risk factors such as aging, genetic susceptibility, environmental risk factors and lifestyle. Since the etiology of AMD is not fully known, it would be essential to identify disease risk factors and novel predictive risk factors to detect AMD at an early stage.

The expression data were obtained from the Gene Expression Omnibus database. Samples were quantile normalized, and log2 transformed. Furthermore, outlier samples were removed by hierarchical clustering. R limma was used to run a linear model and identify differentially expressed genes (DEGs). As a result, 33 genes were discovered with a q-value less than 0.05 and a |log (FC)|≥0.7. With a machine learning (ML) approach, DEGs were applied to discriminate between the case and control samples. Furthermore, FeatureSelect is used to extract the most effective separator genes. Nine genes were identified as the best disease discriminator genes through 11 feature selection algorithms.

The gene set found in the study distinguishes healthy samples from patient samples with an accuracy of 87.5 %. We found DEF119B, UBD, and GRP to be three novel potential AMD candidate biomarkers using ML models and feature selection.

Machine learning can be beneficial in diagnosing, preventing and treating diseases, especially in diseases such as AMD that do not have a clear etiology.

Hypertension is the main reason why the incidence of cardiovascular disease has increased year-by-year and early diagnosis of hypertension is necessary to reducing the incidence of cardiovascular disease. This also puts forward higher requirements for the accuracy of diagnosis. We tried a variety of feature selection methods to improve the accuracy of logistic regression (LR).

We collected 397 samples from Nanjing, Jiangsu, China between Jan 2016 and Dec 2017, including 178 hypertension samples and 219 control samples. It includes not only clinical and laboratory data, but also imaging data. We focused on the difference of imaging attributes between the control group and the hypertension group, and analyzed the correlation coefficients of all attributes. In order to establish the optimal LR model, this study tried three different feature selection methods, including statistical analysis, random forest (RF) and extreme gradient boosting (XGBoost). The area under the ROC curve (AUC) and accuracy were used as the main criterion for model evaluation.

In the prediction of hypertension, the performance of LR with RF as the feature selection method (accuracy: 0.910; AUC: 0.924) was better than the performance of LR with XGBoost as the feature selection method (accuracy: 0.897; AUC: 0.915) and the performance of LR with statistical analysis as the feature selection method (accuracy: 0.872; AUC: 0.926).

LR with RF as the feature selection method may provide accurate results in predicting hypertension. Carotid intima-media thickness (cIMT) and pulse wave velocity at the end of systole (ESPWV) are two key imaging indicators in the prediction of hypertension.

Proper and quick diagnosis of disease is necessary in the medical field for the correct and timely treatment. This issue becomes more important when faced to different diseases with similar symptoms, such as thyroid disease, which has similar symptoms to some disease such as cardiovascular disease. Data mining and machine learning techniques are reliable and valuable methods that can improve the ability of physicians for correctly diagnosis and treatment. The main goal of this research is to extract rules of thyroid disease,

Create the features and analyze feature selection algorithms including filter-based, wrapper based and the genetic algorithm to select the most effective features for thyroid diagnosis. The analysis also performed using decision trees models, random forest, bagging, boosting, and stacking methods for diagnosis and improvement of the illness classes precision that including Hypothyroidism and Hyperthyroidism. Model evaluation was performed with four metrics of accuracy, precision, recall, and F-measure.

This research was conducted on data from the University of California (UCI), which included 7200 records with 21 features. Experimental results showed that the genetic algorithm (GA) has a maximum efficiency in feature selection, and the boosted tree with created features produced maximum F-measure among other classifier.

Mental arithmetic analysis based on Electroencephalogram (EEG) signals can help understand disorders, such as attention-deficit hyperactivity, dyscalculia, or autism spectrum disorder where the difficulty in learning or understanding the arithmetic exists. Most mental arithmetic recognition systems rely on features of a single channel of EEG; however, the relationships between EEG channels in the form of effective brain connectivity analysis can contain valuable information. This study aims to find distinctive, effective brain connectivity features and create a hierarchical feature selection for effectively classifying mental arithmetic and baseline tasks.

We estimated effective connectivity using Directed Transfer Function (DTF), direct DTF (dDTF) and Generalized Partial Directed Coherence (GPDC) methods. These measures determine the causal relationship between different brain areas. A hierarchical feature subset selection method selects the most significant effective connectivity features. Initially, Kruskal–Wallis test was performed. Consequently, five feature selection algorithms, namely, Support Vector Machine (SVM) method based on Recursive Feature Elimination, Fisher score, mutual information, minimum Redundancy Maximum Relevance (RMR), and concave minimization and SVM are used to select the best discriminative features. Finally, the SVM method was used for classification. 

The obtained results indicated that the best EEG classification performance in 29 participants and 60 trials is obtained using GPDC and feature selection via concave minimization method in Beta2 (15-22Hz) frequency band with 89% accuracy. 

This new hierarchical automated system could be helpful in the discrimination of mental arithmetic and baseline tasks from EEG signals effectively.

Cardiovascular disease (CVD) is the first cause of world death, and myocardial infarction (MI) is one of the five primary disorders of CVDs which the patient electrocardiogram (ECG) analysis plays a dominant role in MI diagnosis. This research aims to evaluate some extracted features of ECG data to diagnose MI.

In this paper, we used the Physikalisch‑Technische Bundesanstalt database and extracted some morphological features, such as total integral of ECG, integral of the T‑wave section, integral of the QRS complex, and J‑point elevation from a cycle of normal and abnormal ECG waveforms. Since the morphology of healthy and abnormal ECG signals is different, we applied integral to different ECG cycles and intervals. We executed 100 of iterations on a 10‑fold and 5‑fold cross‑validation method and calculated the average of statistical parameters to show the performance and stability of four classifiers, namely logistic regression (LR), simple decision tree, weighted K‑nearest neighbor, and linear support vector machine. Furthermore, different combinations of proposed features were employed as a feature selection procedure based on classifier’s performance using the aforementioned trained classifiers.

The results of our proposed method to diagnose MI utilizing all the proposed features with an LR classifier include 90.37%, 94.87%, and 86.44% for accuracy, sensitivity, specificity, respectively. Also, we calculated the standard deviation value for the accuracy of 0.006.

Our proposed classification‑based method successfully classified and diagnosed MI using different combinations of presented features. Consequently, all proposed features are valuable in MI diagnosis and are praiseworthy for future works.