Test Data Research Articles

Comparative analysis of machine learning algorithms for phishing website detection

As phishing assaults continue to pose a serious hazard in the digital world, trustworthy detection techniques are required. The effectiveness of machine learning techniques in detecting phishing websites is investigated in this study. The best-performing models were XGBoost and Multilayer Perceptrons (MLPs), which obtained test data accuracy of 90.4% and 90.3%, respectively. On the test data, the Random Forest and Decision Tree models showed competitive accuracies of 86.5% and 87.3%, respectively. SVMs, or support vector machines, performed admirably as well, obtaining an accuracy of 86.4% on the test set. Notably, with accuracy of 74.0% on the test data, the Autoencoder Neural Network demonstrated a restricted level of efficacy. These results highlight the effectiveness of XGBoost and MLPs in precisely detecting phishing websites, offering academics and practitioners in cybersecurity useful information.

Hepatitis C prediction using SVM, logistic regression and decision tree

Hepatitis C is an infection of the liver brought on by the HCV virus. In this condition, early diagnosis is challenging because of the delayed onset of symptoms. Predicting well enough can spare patients from permeant liver damage. The primary goal of this work is to use several machine learning methods to forecast this disease based on widely available and reasonably priced blood test data in order to diagnose and treat patients early on. Three machine learning techniques support vector machine (SVM), logistic regression, decision tree, has been applied on one dataset in this work. To find a suitable approach for illness prediction, the confusion matrix, precision, recall, F1 score, accuracy, receiver operating characteristics (ROC), performances of different strategies have been assessed. The SVM model's overall accuracy is 0.92, the highest among the three models.

Wear particle image analysis: feature extraction, selection and classification by deep and machine learning

Purpose This study aims to explore the integration of machine learning (ML) in tribology to optimize lubrication interval decisions, aiming to enhance equipment lifespan and operational efficiency through wear image analysis. Design/methodology/approach Using a data set of scanning electron microscopy images from an internal combustion engine, the authors used AlexNet as the feature extraction algorithm and the J48 decision tree algorithm for feature selection and compared 15 ML classifiers from the lazy-, Bayes and tree-based families. Findings From the analyzed ML classifiers, instance-based k-nearest neighbor emerged as the optimal algorithm with a 95% classification accuracy against testing data. This surpassed individually trained convolutional neural networks’ (CNNs) and closely approached ensemble deep learning (DL) techniques’ accuracy. Originality/value The proposed approach simplifies the process, enhances efficiency and improves interpretability compared to more complex CNNs and ensemble DL techniques.

Non-linear Kernel Optimisation of Support Vector Machine Algorithm for Online Marketplace Sentiment Analysis

Twitter is a social media platform that is very important in the digital world. Fast communication and interaction make Twitter a vital information center in sentiment analysis. The purpose of this research is to classify public opinion about the presence of marketplaces in Indonesia, both positive and negative sentiments, using a Non-linear SVM algorithm based on 1276 tweets. This research involves the stages of data pre-processing, labeling, feature extraction using TF-IDF, and data division into three scenarios: 80% training data and 20% test data, 50% training data and 50% test data scenario, and 20% training data and 80% test data scenario. The last process, GridSearchCV, combines cross-validation and non-linear SVM parameters for model evaluation using a confusion matrix. The best SVM model resulting from the scenario was 80% training and 20% test data, with hyperparameters Gamma = 100 and C = 0.01, achieving 89% accuracy. When tested on never-before-seen data, the accuracy increased to 90%, with an f1-score of 91%, precision of 88%, and recall of 95% on negative sentiments. In conclusion, evaluating the performance of non-linear SVM kernels with a combination of hyperparameter values can improve accuracy, especially on public response information about online marketplaces and public sentiment.

Oversampling techniques for imbalanced data in regression

Our study addresses the challenge of imbalanced regression data in Machine Learning (ML) by introducing tailored methods for different data structures. We adapt K-Nearest Neighbor Oversampling-Regression (KNNOR-Reg), originally for imbalanced classification, to address imbalanced regression in low population datasets, evolving to KNNOR-Deep Regression (KNNOR-DeepReg) for high-population datasets. For tabular data, we also present the Auto-Inflater neural network, utilizing an exponential loss function for Autoencoders. For image datasets, we employ Multi-Level Autoencoders, consisting of Convolutional and Fully Connected Autoencoders. For such high-dimension data our approach outperforms the Synthetic Minority Oversampling Technique for Regression (SMOTER) algorithm for the IMDB-WIKI and AgeDB image datasets. For tabular data we conducted a comprehensive experiment using various models trained on both augmented and non-augmented datasets, followed by performance comparisons on test data. The outcomes revealed a positive impact of data augmentation, with a success rate of 83.75% for Light Gradient Boosting Method (LightGBM) and 71.57% for the 18 other regressors employed in the study. This success rate is determined by the frequency of instances where models performed better when augmented data was used compared to instances with no augmentation. Access to the comparative code can be found in GitHub.

Direct multi-fidelity integration of 3D CFD models in a gas turbine with numerical zooming method

Multi-fidelity simulation improves the simulation accuracy and captures more detailed information about aero engines under limited computing resources, which is implemented by coupling different levels of models using numerical zooming methods. However, there is an obvious problem in traditional zooming methods such as the iterative coupled zooming method or mini-map method: both the convergence and accuracy depend highly on the component general characteristic maps. Based on the investigation of a micro gas turbine, a direct zooming method (Cycle with CFD in it, CWCFD) is developed. It directly embeds the 3D CFD compressor and turbine model into a 0D component-level model without component general characteristic maps. Then, the CWCFD zooming method is compared with the traditional 0D component-level model in terms of the throttle characteristics of the micro gas turbine, and the experimental data of the ground test is performed to verify the effectiveness of the CWCFD zooming methods. The results indicate that the CWCFD zooming method matches well with the test data better than the traditional 0D component-level model.

The Patient Activation Measure-13 (PAM-13) in an oncology patient population: psychometric properties and dimensionality evaluation

BackgroundAccurate assessment and enhancement of health-related skills among oncology patients are pivotal for optimizing cancer care. The Patient Activation Measure (PAM-13), a questionnaire designed to reflect an individual’s knowledge, skills, and confidence in self-healthcare management, has been validated across diverse countries and settings. Concerns have been raised regarding the cross-situational applicability, as patients with specific diseases and cultural backgrounds interpret questionnaire items differently. This study aimed to examine the structural validity and psychometric properties of the PAM-13 in an oncological patient cohort.MethodsBaseline data from a longitudinal non-randomized controlled study involving cancer out-patients (n = 1,125) from Comprehensive Cancer Centres in Southern Germany were analysed. The German version of the PAM-13 was employed. With classical test and item response theory methods data quality, reliability, convergent and structural validity, as well as psychometric properties were assessed. Exploratory (EFA) and confirmatory factor analyses (CFA) were employed to investigate the postulated unidimensionality of the underlying construct. With a partial credit model (PCM) we examined item fit, targeting, local independence and differential item functioning.ResultsParticipants were predominantly female (73.0%) with a breast cancer diagnosis (41.3%). While items were generally well-accepted, ceiling effects were observed and a high mean PAM-13 score (69.7, SD = 14.2) was noted, potentially compromising responsiveness to interventions. Reliability was adequate (Cronbach’s α = 0.81), person and item separation reliability were good to excellent (0.81 and 0.99, respectively). Explorations of the unidimensionality of the construct (EFA, CFA, PCM) yielded inconclusive results, hinting towards a two-factor solution. Item difficulty rankings deviated from the original. No differential item functioning was identified, and local independence was confirmed.ConclusionsWhile the PAM-13 serves as a valuable instrument for comprehending and promoting health-related skills in cancer patients, the identification of ceiling effects, disordered item-difficulty rankings, and inconclusive findings regarding unidimensionality contribute to the expanding body of evidence, emphasizing the dependency of PAM-13’s validity and reliability on distinctive characteristics within the population under investigation. Future research should prioritize refining or adding PAM-13 items to better capture the specific health-related challenges within diverse populations, paving the way for more effective patient engagement strategies in oncology.Trial registration numberDRKS00021779

Seismic performance of vertical strut and lateral brace suspended continuous‐plasterboard ceiling system with all edges free in the plane of ceiling

AbstractSuspended ceilings are critical nonstructural elements, and their seismic damage in buildings highlighted the incompatibility between the design and performance of structural and nonstructural elements. In order to study the performance of less researched continuous‐plasterboard ceilings, shake table testing was conducted on a ceiling system vertically supported at grid ends with free edges in its plane and suspended by vertical struts and lateral braces. The system had a clearance of 20 mm at the grid ends to accommodate ceiling movements. The ceiling performed satisfactorily for floor accelerations ranging from 0.2 g to 1.4 g without any visible damage. However, the ceiling was slightly rotated and lightly damaged at its perimeter for an extreme dynamic loading of sinusoidal excitation at its natural frequencies. In addition, the experimental performance of the ceiling was numerically validated using nonlinear and linearized responses of sub‐assemblage test data of critical components and connections. It was observed that the developed numerical models can be used to predict the behavior of such ceiling systems as an alternative to evaluation by shake table testing.

Site Identification and Next Choice Protocol for Hit-to-Lead Optimization.

Time efficiency and cost savings are major challenges in drug discovery and development. In this process, the hit-to-lead stage is expected to improve efficiency because it primarily exploits the trial-and-error approach of medicinal chemists. This study proposes a site identification and next choice (SINCHO) protocol to improve the hit-to-lead efficiency. This protocol selects an anchor atom and growth site pair, which is desirable for a hit-to-lead strategy starting from a 3D complex structure. We developed and fine-tuned the protocol using a training data set and assessed it using a test data set of the preceding hit-to-lead strategy. The protocol was tested for experimentally determined structures and molecular dynamics (MD) ensembles. The protocol had a high prediction accuracy for applying MD ensembles, owing to the consideration of protein flexibility. The SINCHO protocol enables medicinal chemists to visualize and modify functional groups in a hit-to-lead manner.

Quantitative radiomics analysis of imaging features in adults and children Mycoplasma pneumonia

PurposeThis study aims to explore the value of clinical features, CT imaging signs, and radiomics features in differentiating between adults and children with Mycoplasma pneumonia and seeking quantitative radiomic representations of CT imaging signs.Materials and methodsIn a retrospective analysis of 981 cases of mycoplasmal pneumonia patients from November 2021 to December 2023, 590 internal data (adults:450, children: 140) randomly divided into a training set and a validation set with an 8:2 ratio and 391 external test data (adults:121; children:270) were included. Using univariate analysis, CT imaging signs and clinical features with significant differences (p < 0.05) were selected. After segmenting the lesion area on the CT image as the region of interest, 1,904 radiomic features were extracted. Then, Pearson correlation analysis (PCC) and the least absolute shrinkage and selection operator (LASSO) were used to select the radiomic features. Based on the selected features, multivariable logistic regression analysis was used to establish the clinical model, CT image model, radiomic model, and combined model. The predictive performance of each model was evaluated using ROC curves, AUC, sensitivity, specificity, accuracy, and precision. The AUC between each model was compared using the Delong test. Importantly, the radiomics features and quantitative and qualitative CT image features were analyzed using Pearson correlation analysis and analysis of variance, respectively.ResultsFor the individual model, the radiomics model, which was built using 45 selected features, achieved the highest AUCs in the training set, validation set, and external test set, which were 0.995 (0.992, 0.998), 0.952 (0.921, 0.978), and 0.969 (0.953, 0.982), respectively. In all models, the combined model achieved the highest AUCs, which were 0.996 (0.993, 0.998), 0.972 (0.942, 0.995), and 0.986 (0.976, 0.993) in the training set, validation set, and test set, respectively. In addition, we selected 11 radiomics features and CT image features with a correlation coefficient r greater than 0.35.ConclusionThe combined model has good diagnostic performance for differentiating between adults and children with mycoplasmal pneumonia, and different CT imaging signs are quantitatively represented by radiomics.

Research on service life prediction and decorative properties of acrylic paint in Vietnam's marine atmosphere by acceleration test

Vietnam is located in a tropical climate zone, especially coastal climates that quickly destroy coating materials that protect steel structures. Therefore, the service life of coatings is always of concern to manufacturers and users. This article presents the results of surveying changes in gloss, color and service life prediction of acrylic paint by accelerated tests, that have simulated according to marine tropical conditions, which are correlated with the atmosphere at Dam Bay station (Hon Tre Island, Nha Trang city, Khanh Hoa province, Vietnam). Gloss is determined according to ISO 2813:2014, and color change is determined according to standard ASTM E308-12. The service life of the acrylic paint system is determined according to standard GOST 9.401-2018. Calculation results based on accelerated test data show that the corrosion time of the acrylic paint system reaches Ri 3 (according to ISO 4628-3:2016 assessment), the service life of acrylic paint Tar 5366 is 2.91 years, and acrylic paint AR-752 is 2.66 years.

Air-conditioning load characteristics and grey box predicting model in subway stations

With the continuous expansion of metro, energy conservation of air-conditioning system in subway stations draws the attention of the whole world. Most of the researches are focused on improving the energy efficiency of the system, while the load characteristics of subway stations are paid little attention. The inaccurate load prediction in the designing of air-conditioning system results in redundancy and low operating efficiency. Therefore, accurately predicting the air-conditioning load is crucial for energy saving in subway stations. In this paper, an innovative three-stage subway air-conditioning load test method was proposed and implemented in nine subway stations in Qingdao. Based on the test data, the special air-conditioning load characteristics of subway stations as underground buildings were analyzed. The results showed that electrical equipment load accounted for the largest part, reaching 53.7 %, while only 42 % of the equipment power was converted into air-conditioning load. The positive role of the envelope should not be ignored, especially in the case of entrance and exit walls, which bore almost all of the fresh air load. Based on the three-stage test, a grey box model was established and the error was validated to be within 6 %. With this model, the load of air-conditioning system can be calculated by inputting passenger flow, power of electrical equipment and public area. This paper provides a reference for the design and operation of air-conditioning system in subway stations.

Contribution to Characterizing the Meat Quality of Protected Designation of Origin Serrana and Preta de Montesinho Kids Using the Near-Infrared Reflectance Methodology

The aims of this study were to describe and compare the meat quality characteristics of male and female kids from the “Serrana” and “Preta de Montesinho” breeds certified as “Cabrito Transmontano” and reinforce the performance of near-infrared reflectance (NIR) spectra in predicting these quality characteristics and discriminating among breeds. Samples of Longissimus thoracis (n = 32; sixteen per breed; eight males and eight females) were used. Breed significantly affected meat quality characteristics, with only color and fatty acid (FA) (C12:0) being influenced by sex. The meat of the “Serrana” breed proved to be more tender than that of the “Preta de Montesinho”. However, the meat from the “Preta de Montesinho” breed showed higher intramuscular fat content and was lighter than that from the “Serrana” breed, which favors its quality of color and juiciness. The use of NIR with the linear support vector machine regression (SVMR) classification model demonstrated its capability to quantify meat quality characteristics such as pH, CIELab color, protein, moisture, ash, fat, texture, water-holding capacity, and lipid profile. Discriminant analysis was performed by dividing the sample spectra into calibration sets (75 percent) and prediction sets (25 percent) and applying the Kennard–Stone algorithm to the spectra. This resulted in 100% correct classifications with the training data and 96.7% accuracy with the test data. The test data showed acceptable estimation models with R2 > 0.99.

Experimental and numerical investigations of S890 and S960 ultra-high strength steel circular hollow section columns

The experimental and numerical investigations of the buckling behaviour and load-carrying capacities of S890 and S960 ultra-high strength steel (UHSS) circular hollow section (CHS) columns were conducted and reported in this paper. The experimental program encompassed tensile coupon tests, initial global and local geometric imperfection measurements and pin-ended column tests. Parallelly, the numerical modelling program was carried out, where the developed finite element (FE) models were validated with reference to the experimental observations and then employed in parametric studies to result in additional numerical data, considering various cross-sectional dimensions and member lengths. Based on the obtained FE and test data, the applicability of the relevant codified design buckling curves in the current European, Australian and American specifications to S890 and S960 UHSS CHS columns was evaluated, since the aforementioned standards only cover the design rules with steel material yield strength no greater than 700 MPa. The assessment results showed that the European code and Australian standard can generally accurately predict the resistances of S890 and S960 UHSS CHS columns, but result in some unsafe data points, while the American specification yielded accurate and consistent resistance predictions. Finally, a revised Eurocode design buckling curve was proposed, accompanied by the reliability assessment through statistical analyses, with the improved design accuracy and consistency revealed.

The symptomatology of Attention-Deficit/Hyperactivity Disorder and the genetic control of vitamin D levels.

Vitamin D deficiency has been associated with psychiatric disorders and behavioral phenotypes such as Attention-Deficit/Hyperactivity Disorder (ADHD). Considering that vitamin D levels are polygenic, we aim to evaluate the overall effects of its genetic architecture on symptoms of inattention, hyperactivity, and impulsivity and on the serum levels of vitamin D in two independent samples of adults, as well as the specific effects of five relevant polymorphisms in vitamin D-related genes. We evaluated 870 subjects from an ADHD sample (407 cases and 463 controls) and 319 subjects from an academic community (nutrigenetic sample). Vitamin D serum levels were obtained through Elisa test and genetic data by TaqMan™ allelic discrimination and Infinium PsychArray-24 BeadChip genotyping. Polygenic Scores (PGS) were calculated on PRSice2 based on the latest GWAS for Vitamin D and statistical analyses were conducted at Plink and SPSS software. Vitamin D PGSs were associated with inattention in the ADHD sample and with hyperactivity when inattention symptoms were included as covariates. In the nutrigenetic sample, CYP2R1 rs10741657 and DHCR7 rs12785878 were nominally associated with impulsivity and hyperactivity, respectively, and both with vitamin D levels. In the clinical sample, RXRG rs2134095 was associated with impulsivity. Our findings suggest a shared genetic architecture between vitamin D levels and ADHD symptoms, as evidenced by the associations observed with PGS and specific genes related to vitamin D levels. Interestingly, differential effects for vitamin D PGS were found in inattention and hyperactivity, which should be considered in further studies involving ADHD.

Intelligent Detection of Surface Defects in High-Speed Railway Ballastless Track Based on Self-Attention and Transfer Learning

The detection of ballastless track surface (BTS) defects is a prerequisite for ensuring the safe operation of high-speed railways. Traditional convolutional neural networks fail to fully exploit contextual information and lack global pixel representations. The extensive stacking of convolutions leads deep learning models to play a black-box detection role, lacking interpretability. Due to the current lack of sufficient high-quality surface data for ballastless tracks, it is a severe constraint on the accurate identification of the substructure state in high-speed railways. This paper proposes an intelligent detection method for BTS defects named TrackNet based on self-attention and transfer learning. The method enhances the fusion ability of global features of BTS defects using multihead self-attention. The model’s dependence on extensive defect data is reduced by transferring knowledge from large-scale publicly available datasets. Experimental results demonstrate that compared to advanced Swin Transformer model results, the TrackNet model achieves improvements in average accuracy and F1-score by 5.15% and 5.16%, respectively, on limited test data. The TrackNet model visualizes the decision regions of the model in identifying BTS defects, revealing the black-box recognition mechanism of deep learning models. This research performs engineering applications and provides valuable insights for the multiclass recognition of BTS defects in high-speed railways.

Enhancing Transparency of Climate Efforts: MITICA’s Integrated Approach to Greenhouse Gas Mitigation

Under the Paris Agreement, countries must articulate their most ambitious mitigation targets in their Nationally Determined Contributions (NDCs) every five years and regularly submit interconnected information on greenhouse gas (GHG) aspects, including national GHG inventories, NDC progress tracking, mitigation policies and measures (PAMs), and GHG projections in various mitigation scenarios. Research highlights significant gaps in the definition of mitigation targets and the reporting on GHG-related elements, such as inconsistencies between national GHG inventories, projections, and mitigation targets, a disconnect between PAMs and mitigation scenarios, as well as varied methodological approaches across sectors. To address these challenges, the Mitigation-Inventory Tool for Integrated Climate Action (MITICA) provides a methodological framework that links national GHG inventories, PAMs and GHG projections, applying a hybrid decomposition approach that integrates machine learning regression techniques with classical forecasting methods for developing GHG emission projections. MITICA enables mitigation scenario generation until 2050, incorporating over 60 PAMs across Intergovernmental Panel on Climate Change (IPCC) sectors. It is the first modelling approach that ensures consistency between reporting elements, aligning NDC progress tracking and target setting with IPCC best practices while linking climate change with sustainable economic development. MITICA’s results include projections that align with observed trends, validated through cross-validation against test data, and employ robust methods for evaluating PAMs, thereby establishing its reliability.

A tightly coupled GNSS RTK/IMU integration with GA-BP neural network for challenging urban navigation

Abstract Intelligent transportation system is increasing the importance of real-time acquisition of positioning, navigation, and timing information from high-accuracy global navigation satellite systems (GNSS) based on carrier phase observations. The complexity of urban environments, however, means that GNSS signals are prone to reflection, diffraction and blockage by tall buildings, causing a degraded positioning accuracy. To address this issue, we have proposed a tightly coupled single-frequency multi-system single-epoch real-time kinematic (RTK) GNSS/inertial measurement unit (IMU) integration algorithm with the assistance of genetic algorithm back propagation based on low-cost IMU equipment for challenging urban navigation. Unlike the existing methods, which only use IMU corrections predicted by machine learning as a direct replacement of filtering corrections during GNSS outages, this algorithm introduces a more accurate and efficient IMU corrections prediction model, and it is underpinned by a dual-check GNSS assessment where the weights of GNSS measurements and neural network predictions are adaptively adjusted based on duration of the integrated system GNSS failure, assisting RTK/IMU integration in GNSS outages or malfunction conditions. Field tests demonstrate that the proposed prediction model results in a 68.69% and 69.03% improvement in the root mean square error in the 2D and 3D component when the training and testing data are collected under 150 s GNSS signal-blocked conditions. This corresponds to 52.43% and 51.27% for GNSS signals discontinuously blocked with 500 s.

IAQ Prediction in Apartments Using Machine Learning Techniques and Sensor Data

This study explores the capability of machine learning techniques (MLTs) in predicting IAQ in apartments. Sensor data from kitchen air monitoring were used to determine the conditions in the living room. The analysis was based on several air parameters—temperature, relative humidity, CO2 concentration, and TVOC—recorded in five apartments. Multiple input–multiple output prediction models were built. Linear (multiple linear regression and multilayer perceptron (MLP)) and nonlinear (decision trees, random forest, k-nearest neighbors, and MLP) methods were investigated. Five-fold cross-validation was applied, where four apartments provided data for model training and the remaining one was the source of the test data. The models were compared using performance metrics (R2, MAPE, and RMSE). The naive approach was used as the benchmark. This study showed that linear MLTs performed best. In this case, the coefficients of determination were highest: R2 = 0.94 (T), R2 = 0.94 (RH), R2 = 0.63 (CO2), R2 = 0.84 (TVOC, based on the SGP30 sensor), and R2 = 0.92 (TVOC, based on the SGP30 sensor). The prediction of distinct indoor air parameters was not equally effective. Based on the lowest percentage error, best predictions were attained for indoor air temperature (MAPE = 1.57%), relative humidity (MAPE = 2.97%RH), and TVOC content (MAPE = 0.41%). Unfortunately, CO2 prediction was loaded with high error (MAPE = 20.83%). The approach was particularly effective in open-kitchen apartments, and they could be the target for its application. This research offers a method that could contribute to attaining effective IAQ control in apartments.

Development of a defined approach for eye hazard identification of solid chemicals according to the three UN GHS categories.

Currently two OECD adopted defined approaches (DA) for eye hazard identification of non-surfactant liquids exist (OECD TG467). The purpose of the current study was to develop a DA for eye hazard identification according to the three UN GHS categories (Cat.1, Cat. 2, No Cat.) for solid chemicals: the DAS. The DAS combines two test methods described in OECD TG437 and TG492. The DAS was developed based on in-depth statistical analysis of a database on solids for which in vitro and historically curated in vivo Draize eye test data exist. The performance of the DAS was assessed by comparing the predictions with the classification based on in vivo Draize eye test data, on the one hand, and with the performance criteria established by the OECD expert group, on the other hand. In a first tier of the DAS, the SkinEthic™ HCE EIT method (TG492) is used to distinguish No Cat. from classified substances. For classified substances the BCOP LLBO method (TG437) is used to identify Cat. 1, the remaining solids are predicted Cat. 2. In summary, 77.4% Cat. 1 (N = 31), 52.3% Cat. 2 (N = 18) and 70.0% of No Cat. (N = 60) solids were correctly identified compared to the classification based on the Draize eye test. The percentage of correct predictions met the minimum OECD established performance values of 75% Cat. 1, 50% Cat. 2, and 70% No Cat. and the percentage of mispredictions was below the established maximum values. Therefore, inclusion of the DAS in OECD TG 467 has been achieved.