Separate Classes Research Articles

Integrating Virtual Reality Into English Language Instruction: Assessing Efficacy And Student Motivation

This study investigates the effectiveness of incorporating Virtual Reality Videos into the English language learning model at Universitas Teknologi Digital Indonesia. Employing a quasi-experimental design, the researcher implemented two distinct learning methods across separate class groups. One group utilized the Virtual Reality learning model, while the other adhered to traditional instructional methods. The assessment of the learning model's efficacy employed statistical tests and questionnaire analyses administered to participants post-learning sessions, coupled with classroom observations. Results from the research revealed that students utilizing Virtual Reality videos achieved higher scores compared to those using conventional videos. While questionnaire responses indicated relatively comparable outcomes, Virtual Reality videos demonstrated a marginal advantage concerning student learning experiences and motivation. This study contributes valuable insights into the pedagogical benefits of integrating VR technology, emphasizing its positive impact on language acquisition within the higher education context

Dual intent view contrastive learning for knowledge aware recommender systems

Knowledge-aware recommendation systems often face challenges owing to sparse supervision signals and redundant entity relations, which can diminish the advantages of utilizing knowledge graphs for enhancing recommendation performance. To tackle these challenges, we propose a novel recommendation model named Dual-Intent-View Contrastive Learning network (DIVCL), inspired by recent advancements in contrastive and intent learning. DIVCL employs a dual-view representation learning approach using Graph Neural Networks (GNNs), consisting of two distinct views: a local view based on the user-item interaction graph and a global view based on the user-item-entity knowledge graph. To further enhance learning, a set of intents are integrated into each user-item interaction as a separate class of nodes, fulfilling three crucial roles in the GNN learning process: (1) providing fine-grained representations of user-item interaction features, (2) acting as evaluators for filtering relevant relations in the knowledge graph, and (3) participating in contrastive learning to strengthen the model’s ability to handle sparse signals and redundant relations. Experimental results on three benchmark datasets demonstrate that DIVCL outperforms state-of-the-art models, showcasing its superior performance. The implementation is available at: https://github.com/yzxx667/DIVCL.

Assessment of using transfer learning with different classifiers in hypodontia diagnosis

BackgroundHypodontia is the absence of one or more teeth in the primary or permanent dentition during development, and radiographic imaging is the most common method of diagnosis. However, in recent years, artificial intelligence-based decision support systems have been employed to make highly accurate diagnoses. The aim of this study was to classify single premolar agenesis, multiple premolar agenesis, and without tooth agenesis using various artificial intelligence approaches.MethodsOne thousand sixty-eight panoramic radiographs from pediatric patients aged between 6 and 12 years without systemic disease were sorted into three separate classes: single premolar agenesis (n = 336), multiple premolar agenesis (n = 324), and without tooth agenesis (n = 408). Pretrained convolutional neural network models (AlexNet, DarkNet-19, DarkNet-53, DenseNet-201, EfficientNet, GoogLeNet, InceptionV3, IncResV2, MobileNetV2, NasNet-Mobile, Places365, ResNet-18, ResNet-50, ResNet-101, ShuffleNet, SqueezeNet, VGG-16, VGG-19, and Xception) were used for training with the fine-tuning method and different machine learning classifiers (decision trees, discriminant analysis, logistic regression, naive Bayes, support vector machines, nearest neighbor, ensemble method, and artificial neural network). The dataset was divided into 80% for training and 20% for testing. Performance was evaluated via accuracy, recall, precision, F1-score, specificity and area under the curve (AUC) parameters.ResultsAll of the data were classified via a VGG-19 model with a bilayered neural network classifier, which achieved 95.63% accuracy, 93.26% precision, 93.34% recall, 96.73% specificity, 93.25% F1-score and 95.03% AUC and was identified as the most successful model. The accuracy values for this model were distributed as follows: 96.72% for without tooth agenesis, 95.79% for multiple premolar agenesis, and 94.39% for single premolar agenesis.ConclusionsSuccessful results of pretrained models have been demonstrated for the radiographic diagnosis of hypodontia in pediatric patients. It is expected that artificial intelligence approaches will facilitate the diagnosis of hypodontia.

Combining parametric discreteness and continuity for analyzing dynamic models of structures

This paper describes a method for building dynamic models that contain rod bending elements with distributed and concentrated inertial and stiffness parameters, and their analysis based on the harmonic element method. As a rule, the vibration effects of structures are calculated on the basis of mass discretization, although the application of such methods entails certain difficulties.Discrete models are considered to be a priori approximations with limited possibilities of error estimation. The dynamic parameters of the model vary depending on its dimensionality as well as on the transformation methods. Numerical results with arrays and matrices of high dimensionality make it difficult to analyze and evaluate the calculation results. Therefore, structural calculations for stationary dynamic effects based on the use of elements with distributed and concentrated masses prevent the above-mentioned consequences of full discretization. However, such discrete-continuum (hybrid) dynamic models require the sewing of heterogeneous elements at the formation stage. In addition, some complications occur when solving these combined systems containing ordinary differential equations and partial differential equations. These issues can be solved by using the author's harmonic element method, implementing the nodal sewing of heterogeneous elements, as well as providing solutions as amplitudes of oscillations of the combined model nodes along certain necessary directions. The specified features of the proposed method allow us to identify it as a separate class with the name of the harmonic element method.

Machine learning applied to arbovirus genome classification

The aim of this study was to evaluate the use of machine learning for rapid classification of arbovirus genomes. Initially, genomic sequences of 17 distinct arboviruses were collected from the National Center for Biotechnology Information database. Genomic sequences of arthropod-specific virus were also collected to compose a separate class, representing a “non-arbovirus” group. Subsequently, the sequences were transformed into canonical k-mer frequencies and used to train supervised classification algorithms such as multinomial logistic regression, decision tree, k-nearest neighbors, support vector machine and multilayer perceptron. Six distinct k-mer values within a range of 1 to 6 were also evaluated. Using 10-fold cross-validation as an evaluation method, the supervised model created with multilayer perceptron and k-mer value 6 presented the best average accuracy (98.8%). In order to evaluate the generalization capacity of the best model obtained, classifications were made using genomic sequences not present in the training database. The results of the classifications generated were evaluated by the metrics of accuracy, precision, recall and f1-score, obtaining values of 98.5%, 98.3%, 98.2% and 98.2%, respectively. Finally, the best model obtained was incorporated into a web application, which allows the input of virus genomic sequences and their classification. The application is freely available for use at https://arbovirusclassifiercanonicalkmer-8fndyh3tsxrftmr66jmpas.streamlit.app.

A Large-Scale Sensitivity Analysis on Latent Embeddings and Dimensionality Reductions for Text Spatializations.

The semantic similarity between documents of a text corpus can be visualized using map-like metaphors based on twodimensional scatterplot layouts. These layouts result from a dimensionality reduction on the document-term matrix or a representation within a latent embedding, including topic models. Thereby, the resulting layout depends on the input data and hyperparameters of the dimensionality reduction and is therefore affected by changes in them. Furthermore, the resulting layout is affected by changes in the input data and hyperparameters of the dimensionality reduction. However, such changes to the layout require additional cognitive efforts from the user. In this work, we present a sensitivity study that analyzes the stability of these layouts concerning (1) changes in the text corpora, (2) changes in the hyperparameter, and (3) randomness in the initialization. Our approach has two stages: data measurement and data analysis. First, we derived layouts for the combination of three text corpora and six text embeddings and a grid-search-inspired hyperparameter selection of the dimensionality reductions. Afterward, we quantified the similarity of the layouts through ten metrics, concerning local and global structures and class separation. Second, we analyzed the resulting 42 817 tabular data points in a descriptive statistical analysis. From this, we derived guidelines for informed decisions on the layout algorithm and highlight specific hyperparameter settings. We provide our implementation as a Git repository at hpicgs/Topic-Models-and-DimensionalityReduction-Sensitivity-Study and results as Zenodo archive at DOI:10.5281/zenodo.12772898.

A step towards quantifying, modeling and exploring uncertainty in biomedical knowledge graphs

Objective:This study aims at automatically quantifying and modelling the uncertainty of facts in biomedical knowledge graphs (BKGs) based on their textual supporting evidence using deep learning techniques. Materials and Methods:A sentence transformer is employed to extract deep features of sentences used to classify sentence factuality using a naive Bayes classifier. For each fact and its supporting evidence in a source KG, the deep feature extractor and the classifier are used to quantify the factuality of each sentence which are then transformed to numerical values in [0,1] before being averaged to get the confidence score of the fact. Results:The fact classification feature extractor enhances the separability of classes in the embedding space. This helped the fact classification model to achieve a better performance than existing factuality classification with hand-crafted features. Uncertainty quantification and modelling were demonstrated on SemMedDB by creating USemMedDB, showing KGB2U’s ability to process large BKGs. A subset of USemMedDB facts is modelled to demonstrate the correlation between the structure of the uncertain BKG and the confidence scores. The best-trained model is used to predict confidence scores of existing and unseen facts. The top-ranked unseen facts were grounded using scientific evidence showing KGB2U’s ability to discover new knowledge. Conclusion:Supporting literature of BKG facts can be used to automatically quantify their uncertainty. Additionally, the resulting uncertain biomedical KGs can be used for knowledge discovery. BKG2U interface and source code are available at http://biofunk.datanets.org/ and https://github.com/BahajAdil/KBG2U respectively.

Early Postpartum Metabolic Heterogeneity Among Women Who Progressed to Type 2 Diabetes After Gestational Diabetes: A Prospective Cohort.

Gestational diabetes mellitus (GDM) poses a significant risk for developing type 2 diabetes mellitus (T2D) and exhibits heterogeneity. However, understanding the link between different types of post-GDM individuals without diabetes and their progression to T2D is crucial to advance personalised medicine approaches. We employed a discovery-based unsupervised machine learning clustering method to generate clustering models for analysing metabolomics, clinical, and biochemical datasets. For this analysis, we selected 225 women who later developed T2D during the 12-year follow-up period from the cohort of 1010 women who returned to a non-diabetic state at 6-9weeks (study baseline) after a GDM pregnancy based on 2-h 75g research OGTTs. The optimal model was selected by assessing Bayesian Information Criterion values, class separation performance, and the potential for clinically distinguishable clusters, accounting for participant prenatal and early postpartum characteristics. The selected model comprises three clusters: pancreatic beta cell dysfunction (cluster-β: median HOMA-B 161.3 and median HOMA-IR 3.8), insulin-resistance (cluster-IR: median HOMA-B 630.5 and median HOMA-IR 16.8), and a mixed cluster (cluster-mixed: median HOMA-B 307.2 and median HOMA-IR 8.6). These clusters are distinguishable based on postpartum blood test parameters such as glucose tolerance, HOMA indices, and fasting lipid profiles including triglycerides, leptin, HDL-c, and adiponectin, as well as participant age and BMI. Metabolomic analysis identified unique molecular signatures for each cluster. However, the time to T2D onset was not statistically significant among the three clusters (p=0.22). This study enhances our understanding of the heterogeneity of early postpartum metabolic profiles that characterise the future onset of T2D diabetes in a diverse cohort of women with GDM, revealing insights into distinct mechanisms and personalised intervention strategies for the prevention of T2D.

Differentiation of the chronic lymphocytic leukemia response to ibrutinib and acalabrutinib treatment by single-cell MALDI-TOF MS imaging.

Ibrutinib and acalabrutinib, Bruton's tyrosine kinase inhibitors (BTKi) used for chronic lymphocytic leukemia (CLL) treatment, aim the same target but their off-target effects are different. The aim of this study was to use single-cell MALDI TOF mass spectrometry imaging to compare the CD19+lymphocytes' mass spectra in untreated and ibrutinib- or acalabrutinib-treated subjects in order to better understand the therapeutic effect of BTKi. 180 cells from 9 male subjects divided in 3 groups (untreated, ibrutinib-treated and acalabrutinib-treated) were analyzed using MALDI-TOF mass spectrometry analyzer. Mass spectra were acquired in the 300-600 Da mass range. Partial least squares discriminant analysis (PLS-DA) was used for evaluation of mass spectra, while Volcano plots and Venn diagram were used for a review of significantly altered m/z values. Cross-validated PLS-DA classification accuracy of cells was 85 %. Ibrutinib-treated cells overlap with the untreated cell population, while acalabrutinib-treated cells form a separate class. 13 m/z signals were specific for each BTKi group. In conclusion, single-cell MALDI-TOF mass spectrometry imaging detects differences in the mass spectra of CD19+ lymphocytes treated with two different BTKi. Drug-specific m/z signal clusters can be identified as a chemical fingerprint of the BTKi effect and represent candidates for the therapeutic response biomarkers.

Blockchain-Enabled Security Framework against Ransomware Attacks Using Machine Learning

In this presentable paper, we first describe and analyze the background of a certain discipline by reviewing the recent scientific literature on this subject and summarizing it about issues of the new research. Ransomware through reasoning can also be considered the last significant stage for cyber extortion blocking access to the resources of the target organization until submission by the latter to certain coercion or payment. There is a separate class of malware known as ransomware. When a computer or some other device suffers a ransomware malware attack, such device is either locked/held hostage or the data within the device is encrypted. Ransom demands (usually a small sum) are placed on the victims for providing the data order for the data translation in the form of a decryption key. Due to these attacks, surveillance means and protective programs are recommended given the prevention of ransomware epidemic outbreaks. The most vulnerable targets are probably those classed as organizations, such as financial institutes and healthcare sectors. Blockchain technology prevents tempering which makes it much more effective than the traditional centralized approach of data storage. Such aspects of blockchain technology can enhance the security perimeter for the detection and prevention of ransomware attacks even more. The objectives of the research are to demonstrate the extent of the problem and to show how problems are identifiable within datasets, which is through the application of machine learning. In this paper, we propose a new security framework that applies machine learning to prevent ransomware attacks and is based on the principles of blockchain technology.

Обнаружение сейсмических событий на основе искусственной нейронной сети SeisDetNet. Часть 1. Архитектура нейронной сети

An artificial neural network, SeisDetNet, has been developed for distinguishing seismic events from seismic noise based on waveform records. The international STEAD database, which contains minute-long records of local earthquakes and seismic noise, was used as the data source for training, validation, and testing. Specifically, we selected waveforms from 27 seismic stations located in Kyrgyzstan and surrounding areas from the KRNET, KNET, KZ, G, and TJ networks, present in the STEAD database. The model architecture is a combination of a convolutional network, designed to extract key features for class separation, and a fully connected network for the task of classifying the input record as either a seismic event or seismic noise. Evaluating the model on the test set showed good results, with the following binary classification metrics: accuracy of 0.983, precision of 0.989, recall of 0.982, and F1-score of 0.985.

ОБОЛОЧКИ, ОБОЛОЧЕЧНЫЕ СТРУКТУРЫ И КРИВОЛИНЕЙНЫЕ СООРУЖЕНИЯ В ФОРМЕ АЛГЕБРАИЧЕСКИХ ПОВЕРХНОСТЕЙ 2-го ПОРЯДКА

Geometricians, architects, and engineers used the real second order surfaces always. In a presented review paper, interesting information and results of research published at the last 20 years and devoted to strength, stability, and vibration of thin and thick, one-layered and many-layered shells, made of different constructive materials, were collected. Illustrations of erections and products of the last years are presented. The real surfaces of the second order form the separate class but all of them can be distributed between other classes of surfaces. It was proved in a paper. The presented materials show that interest of researchers for considered surfaces, shell structures, shells, and curvilinear erections outlined on the second order surfaces does not go out. In this paper, the information containing in 67 used references is exploited.

BIPE: A Bi-Layer Predictive Ensemble Framework for Forest Fire Susceptibility Mapping in Germany

Forest fires diminish forests’ ecological services, including carbon sequestration, water retention, air cooling, and recreation, while polluting the environment and endangering habitats. Despite considerable economic advancements, firefighting strategies remain less than optimal. This paper introduces the Bi-layer Predictive Ensemble (BIPE), an innovative machine learning model designed to enhance the accuracy and generalization of forest fire susceptibility mapping. BIPE integrates model-centric and data-driven strategies, employing automated methods such as 10-fold cross-validation and meta-learning to improve stability and generalization. During its 10-fold cross-validation, BIPE demonstrated excellent performance, with the Area Under the Curve (AUC) values ranging from 0.990 to 0.996 and accuracy levels consistently high, around 97%, underscoring its robust class separation ability and strong generalization across different datasets. Our results confirm that BIPE outperforms traditional high-performance models like Support Vector Machine (SVM), Multilayer Perceptron (MLP), Extreme Gradient Boosting (XGBoost), Deep Neural Network (DNN), and Convolutional Neural Network (CNN), showcasing its practical effectiveness and reliability on the data of nonlinear, high-dimensional, and complex interactions. Additionally, our forest fire susceptibility maps offer valuable complementary information for German forest fire management authorities, enhancing their ability to assess and manage fire risks more effectively.

Material Hardship, Forced Displacement, and Negative Health Outcomes Among Unhoused People Who Use Drugs in Los Angeles, California and Denver, Colorado: A Latent Class Analysis.

Homelessness is a growing concern in the United States, especially among people who use drugs (PWUD). The degree of material hardship among this population may be linked to worse health outcomes. PWUD experiencing homelessness in urban areas are increasingly subjected to policies and social treatment, such as forced displacement, which may worsen material hardship. It is critical to describe hardship among PWUD and examine if it is linked to health outcomes. Data were collected as part of a prospective cohort study of PWUD in Los Angeles, California and Denver, Colorado (n = 476). Analysis sample size was smaller (N = 395) after selecting for people experiencing homelessness and for whom data were complete. Five indicators assessing hardship (difficulty finding food, clothing, restrooms, places to wash/shower, and shelter) in the past three months were obtained from participants at baseline and were used in latent class analysis (LCA). We chose a base latent class model after examination of global fit statistics. We then built three auxiliary models using the three-step Bolck-Croon-Hagenaars (BCH) method to test the relationship of latent class membership to several hypothesized social and health variables in this same three month time period. Fit statistics, minimum classification probabilities, and ease of interpretation indicated a three-class solution for level of material difficulty. We termed these classes "High Difficulty" (n = 82), "Mixed Difficulty" (n = 215), and "Low Difficulty" (n = 98). Average classification probabilities indicated good class separability. "High Difficulty" participants had high probabilities of usually having difficulty accessing all five resources. "Mixed Difficulty" participants indicated a range of difficulty accessing all resources, with restrooms and bathing facilities being the most difficult. "Low Difficulty" participants were defined by high probabilities of never having access difficulty. In auxiliary analyses, there were significant (p < 0.05) differences in experiences of displacement, opioid withdrawal symptoms, nonfatal overdose, and violent victimization between classes. This LCA indicates that among PWUD experiencing homelessness there exist distinct differences in resource access and material hardship, and that these differences are linked with political, social, substance use, and other health outcomes. We add to the literature on the relationship between poverty and health among PWUD. Policies which increase difficulty accessing necessary material resources may negatively impact health in this population.

Hybrid Deep Learning Model for Fault Diagnosis in Centrifugal Pumps: A Comparative Study of VGG16, ResNet50, and Wavelet Coherence Analysis

Significant in various industrial applications, centrifugal pumps (CPs) play an important role in ensuring operational efficiency, yet they are susceptible to faults that can disrupt production and increase maintenance costs. This study proposes a robust hybrid model for accurate fault detection and classification in CPs, integrating Wavelet Coherence Analysis (WCA) with deep learning architectures VGG16 and ResNet50. WCA is initially applied to vibration signals, creating time–frequency representations that capture both temporal and frequency information, essential for identifying subtle fault characteristics. These enhanced signals are processed by VGG16 and ResNet50, each contributing unique and complementary features that enhance feature representation. The hybrid approach fuses the extracted features, resulting in a more discriminative feature set that optimizes class separation. The proposed model achieved a test accuracy of 96.39%, demonstrating minimal class overlap in t-SNE plots and a precise confusion matrix. When compared to the ResNet50-based and VGG16-based models from previous studies, which reached 91.57% and 92.77% accuracy, respectively, the hybrid model displayed better classification performance, particularly in distinguishing closely related fault classes. High F1-scores across all fault categories further validate its effectiveness. This work underscores the value of combining multiple CNN architectures with advanced signal processing for reliable fault diagnosis, improving accuracy in real-world CP applications.

Complexity, interpretability and robustness of GP-based feature engineering in remote sensing

Feature engineering is a crucial step in machine learning that provides better data for the learning algorithms to induce robust models, and this effort should be adapted to the capabilities of each algorithm. For example, classifiers that do not perform data transformations (e.g., cluster-based) perform better when the different classes are separated, typically requiring preprocessed data. Other models (e.g., decision trees) can perform several splits in the feature space, easily obtaining perfect results in training data, but have a higher risk of overfitting with unprocessed data. We use the rbd-GP and M3GP genetic programming algorithms to induce new features based on the original features, to be used by shallow and deep decision tree and random forest models. M3GP is wrapped around a learning algorithm, using its performance as fitness. This way, the induced features are adapted to the classifier, allowing us to compare the complexity of the features induced for the different classifiers. We measure the complexity of the induced features using several structural and functional complexity metrics found in the literature, also proposing a new metric that measures the separability of classes in the feature space. Like other authors, we use complexity as an interpretability metric, selecting three models to discuss and validate based on their performance and size. We apply these methods to remote sensing classification problems and solve two tasks that are hard due to the high similarity between the land cover classes: detecting cocoa agroforest and forecasting forest degradation up to one year in the future.

Genome-wide identification, classification, and expression profiling of the aldehyde dehydrogenase gene family in pepper.

Pepper (Capsicum annuum L.) is one of the most significant vegetable crops worldwide which is known for its pungency and nutritional value. The aldehyde dehydrogenase (ALDH) superfamily encompasses enzymes critical for the detoxification of toxic aldehydes into non-toxic carboxylic acids. A comprehensive genome-wide approach in pepper identified a total of 27 putative ALDH genes grouped into ten families based on the criteria of the ALDH gene nomenclature committee. Both segmental and tandem duplication assisted in the enhancement of CaALDH gene family members. The identified CaALDH members were found to be more closely related to the dicot plants, however, the members were distributed across the phylogenetic tree suggesting the pre-eudicot-monocot separation of the ALDH superfamily members. The gene structure and protein domain were found to be mostly conserved in separate phylogenetic classes, indicating that each family played an important role in evolution. Expression analysis revealed that CaALDHs were expressed in various tissues, developmental stages, and in response to abiotic stresses, indicating that they can play roles in plant growth, development, and stress adaptation. Interestingly, the majority of the CaALDH genes were found to be highly responsive to salinity stress, and only the CaALDH11A1 transcript showed upregulation in cold stress conditions. The presence of cis-acting elements in the promoter region of these genes might have a significant role in abiotic stress tolerance. Overall, these findings add to the current understanding, evolutionary history, and contribution of CaALDHs in stress tolerance, and smooth the path of further functional validation of these genes.

NETWORK INTRUSION DETECTION MODEL BASED ON CONVOLUTIONAL NEURAL NETWORKS AND TABULAR DATA CONVERTED INTO IMAGES

The object of the study is the process of identifying the state of a computer systems and network. The subject of the study are the methods of identifying the state of computer systems and networks. The purpose of this paper is to improve the quality of detecting intrusions into computer networks. The UNSW-NB 15 set, which contains information about the normal functioning of the network and during synthetic intrusions, was used as input. Deep neural networks (DL), their advantages and problems in big data processing are considered. It was found that deep neural networks when processing tabular data require their transformation. Modern methods of tabular data transformation were studied. The results obtained. A method of converting tabular data into an image is proposed. The method converts each object of a separate class from a set of tabular data into an image by mapping the attribute values onto a two-dimensional plane. The method was implemented programmatically using the GOOGLE COLAB cloud service based on Jupyter Notebook. Conclusions. It was found that the use of the proposed conversion method of tabular data into an image made it possible to use a classification model based on the CNN neural network and increase the quality of detection of intrusions into computer networks up to 4%.

Four-Dimensional Lipidomic Analysis Using Comprehensive Online UHPLC × UHPSFC/Tandem Mass Spectrometry.

Multidimensional chromatography offers enhanced chromatographic resolution and peak capacity, which are crucial for analyzing complex samples. This study presents a novel comprehensive online multidimensional chromatography method for the lipidomic analysis of biological samples, combining lipid class and lipid species separation approaches. The method combines optimized reversed-phase ultrahigh-performance liquid chromatography (RP-UHPLC) in the first dimension, utilizing a 150 mm long C18 column, with ultrahigh-performance supercritical fluid chromatography (UHPSFC) in the second dimension, using a 10 mm long silica column, both with sub-2 μm particles. A key advantage of employing UHPSFC in the second dimension is its ability to perform ultrafast analysis using gradient elution with a sampling time of 0.55 min. This approach offers a significant increase in the peak capacity. Compared to our routinely used 1D methods, the peak capacity of the 4D system is 10 times higher than RP-UHPLC and 18 times higher than UHPSFC. The entire chromatographic system is coupled with a high-resolution quadrupole-time-of-flight (QTOF) mass analyzer using electrospray ionization (ESI) in both full-scan and tandem mass spectrometry (MS/MS) and with positive- and negative-ion polarities, enabling the detailed characterization of the lipidome. The confident identification of lipid species is achieved through characteristic ions in both polarity modes, information from MS elevated energy (MSE) and fast data-dependent analysis scans, and mass accuracy below 5 ppm. This analytical method has been used to characterize the lipidomic profile of the total lipid extract from human plasma, which has led to the identification of 298 lipid species from 16 lipid subclasses.

Improving Churn Detection in the Banking Sector: A Machine Learning Approach with Probability Calibration Techniques

Identifying and reducing customer churn have become a priority for financial institutions seeking to retain clients. Our research focuses on customer churn rate analysis using advanced machine learning (ML) techniques, leveraging a synthetic dataset sourced from the Kaggle platform. The dataset undergoes a preprocessing phase to select variables directly impacting customer churn behavior. SMOTETomek, a hybrid technique that combines oversampling of the minority class (churn) with SMOTE and the removal of noisy or borderline instances through Tomek links, is applied to balance the dataset and improve class separability. Two cutting-edge ML models are applied—random forest (RF) and the Light Gradient-Boosting Machine (LGBM) Classifier. To evaluate the effectiveness of these models, several key performance metrics are utilized, including precision, sensitivity, F1 score, accuracy, and Brier score, which helps assess the calibration of the predicted probabilities. A particular contribution of our research is on calibrating classification probabilities, as many ML models tend to produce uncalibrated probabilities due to the complexity of their internal mechanisms. Probability calibration techniques are employed to adjust the predicted probabilities, enhancing their reliability and interpretability. Furthermore, the Shapley Additive Explanations (SHAP) method, an explainable artificial intelligence (XAI) technique, is further implemented to increase the transparency and credibility of the model’s decision-making process. SHAP provides insights into the importance of individual features in predicting churn, providing knowledge to banking institutions for the development of personalized customer retention strategies.