Classification Model Research Articles

Data Transformation and Predictive Analytics of Cardiovascular Disease Using Machine and Ensemble Learning Techniques

About one person dies every minute from cardiovascular disease; consequently, it has almost surpassed war as the largest cause of death in the twenty-first century. In cardiology, early and accurate diagnosis of heart illness is a cornerstone of effective healthcare. Predictive analytics, which involves machine-learning algorithms, can be a great option for contributing towards the early detection of cardiovascular disease. This study evaluates the data preprocessing techniques involved in building machine learning models to predict cardiovascular disease and identify the features contributing to the cardio attack. A novel data transformation technique named the superlative boundary binning method was proposed to enhance machine learning and ensemble learning classification models for predicting cardiac illness based on independent physiological feature parameters. The results revealed that the ensemble learning classifier AdaBoost using the superlative boundary binning method has performed well with a classification accuracy of 93% when compared with the other data transformation and machine learning classifier models.

Modelling Misinformation in Swahili-English Code-switched Texts

Code-switching, which is the mixing of words or phrases from multiple, grammatically distinct languages, introduces semantic and syntactic complexities to sentences which complicate automated text classification. Despite code-switching being a common occurrence in informal text-based communication among most bilingual or multilingual users of digital spaces, its use to spread misinformation is relatively less explored. In Kenya, for instance, the use of code-switched Swahili-English is prevalent on social media. Our main objective in this paper was to systematically re- view code-switching, particularly the use of Swahili-English code-switching to spread misinformation on social media in the Kenyan context. Additionally, we aimed at pre-processing a Swahili-English code-switched dataset and developing a misinformation classification model trained on this dataset. We discuss the process we took to develop the code- switched Swahili-English misinformation classification model. The model was trained and tested using the PolitiKweli dataset which is the first Swahili-English code-switched dataset curated for misinformation classification. The dataset was collected from Twitter (now X) social media platform, focusing on text posted during the electioneering period of the 2022 general elections in Kenya. The study experimented with two types of word embeddings - GloVe and FastText. FastText uses character n-gram representations that help generate meaningful vectors for rare and unseen words in the code-switched dataset. We experimented with both the classical machine learning algorithms and deep learning algo- rithms. Bidirectional Long Short-Term Memory Networks (BiLSTM) algorithm showed the best performance with an f-score of 0.89. The model was able to classify code-switched Swahili-English political misinformation text as fake, fact or neutral. This study contributes to recent research efforts in developing language models for low-resource languages.

Leveraging Vision Transformers for Early Detection of Alzheimer's disease Through MRI Image Analysis

Alzheimer's disease (AD) poses challenges for early recognition due to subtle and overlapping symptoms, making timely care and intervention difficult. This work presents a new method for identifying Alzheimer's disease by combining magnetic resonance imaging (MRI) data with a Vision Transformer (ViT) model. In this research, we develop a method that leverages MRI scans to classify Alzheimer's disease (AD) into four distinct stages: Non-Demented, Very Mild Demented, Mild Demented, and Moderate Demented. To improve the identification of clinical and structural alterations in the brain linked to AD, sophisticated characteristics are retrieved, including voxel-intensity patterns, grey matter volume, and cortical thickness. These characteristics allow for a more accurate representation of MRI data when paired with sophisticated processing methods. Additionally, even in situations where data is scarce, transfer learning is used to maximize the Vision Transformer (ViT) model's classification accuracy, especially when it comes to differentiating between neighboring stages of dementia. The ViT model, a cutting-edge deep learning technique, shows strong performance and dependability in identifying AD phases. Our findings highlight its effectiveness, showcasing a significant improvement in diagnostic accuracy compared to existing methods. The accuracy with which the ViT model can distinguish between different phases of AD emphasizes both its suitability for handling complicated MRI data and its promise for early diagnosis and detection.This paper offers a promising method for enhancing MRI image processing for Alzheimer's disease diagnosis through sophisticated feature extraction, transfer learning, and the use of Vision Transformers (ViTs). Present authors believe this method successfully tackles the difficulties of early and precise detection, making it essential for appropriate medical image analysis.

Hyperconnectivity and Connectome Gradient Dysfunction of Cerebello-Thalamo-Cortical Circuitry in Alzheimer's Disease Spectrum Disorders.

Cerebellar functional connectivity changes have been reported in Alzheimer's disease (AD), but a comprehensive framework integrating these findings is lacking. This retrospective study investigates the cerebello-thalamo-cortical (CTC) circuit in AD, using functional gradient analysis to elucidate deficits and potential biomarkers. We analyzed data from 246 participants enrolled in the Alzheimer's Disease Neuroimaging Initiative (ADNI-3; NCT02854033), including 58 with AD, 103 with mild cognitive impairment (MCI), and 85 cognitively normal (CN) controls, matched for age and sex. All individuals underwent comprehensive neuropsychological assessments (MMSE, MoCA, ADAS-Cog) and MRI scans. We extracted mean time series for 270 brain regions (an extended Power atlas) and computed pairwise functional connectivity, focusing on CTC circuitry. Thalamic and cerebellar connectivity gradients were derived using voxel-wise correlation matrices and the BrainSpace toolbox, defining thalamic and cerebellar masks from the Melbourne subcortical atlas and AAL atlas, respectively. ANCOVA with post hoc analyses, controlling for age and sex, was conducted to assess abnormal CTC connectivity across AD, MCI, and CN groups. LASSO regression identified edges within the CTC circuitry that significantly differed between AD and CN, MCI and CN, AD and MCI, as well as was used to construct Logistic classification model. Pearson correlations were performed to examine relationships between mean CTC connectivity, individual edges, and cognitive scores (MMSE, MoCA, ADAS-Cog). To explore the hierarchical organization of the thalamus and cerebellum, global gradient distributions were compared across groups using two-sample Kolmogorov-Smirnov tests. Additionally, ANCOVA was applied to compare subfield- and functional-level gradients of the thalamus and cerebellum among AD, MCI, and CN. False discovery rate (FDR) corrections were used, setting the statistical significance threshold was set at P < 0.05. AD and MCI individuals exhibited increased CTC connectivity compared to CN (all P < 0.05). Average CTC connectivity did not correlate with cognitive scores (P > 0.05), but specific CTC edges were correlated. LASSO regression identified 20 discriminative edges, achieving high accuracy in AD-CN classification (AUC = 0.92 training, AUC = 0.80 test). Thalamic and cerebellar gradient distributions differed significantly across groups (all P < 0.05), with specific regions showing distinct gradient scores. Five cerebellar functional networks exhibited decreased gradient scores. Significant CTC hyperconnectivity in AD and MCI compared with CN suggests early thalamic and cerebellar dysregulation. Classification analyses effectively distinguished AD vs. CN but were moderate for MCI vs. CN and limited for MCI vs. AD. Gradient analyses revealed global- and subfield-level disruptions in AD, emphasizing the role of thalamic and cerebellar interactions in cognitive decline and offering potential diagnostic markers and therapeutic targets.

Skin microbiome differences in pancreatic adenocarcinoma, other cancers, and healthy controls: a pilot study

IntroductionMany studies have reported the importance of the human microbiome in relationship to the overall health of its host. While recent studies have explored the microbiome’s role in various types of cancer compared to healthy patients, this pilot study is the first to investigate differences in the skin microbiome composition among pancreatic adenocarcinoma patients, individuals with other cancers, and cancer-free controls.MethodsThe study characterizes the skin microbiome’s potential associations with cancer status by analyzing skin swabs from the forehead and cheek of 58 participants using Next Generation Sequencing (NGS), differential abundance analysis, and machine learning techniques.ResultsThe study results indicated that the cancer group displayed a significantly higher mean alpha diversity compared to the control group. Additionally, a machine learning classification model achieved a mean F1 Score of 0.943 in predicting cancer status, indicating measurable differentiation in the skin microbiome between the study groups. This differentiation is supported by differential abundance methods, including ANCOM-BC and MaAsLin2.DiscussionThis pilot study suggests that skin microbiome profiling could serve as a non-invasive biomarker for cancer detection and monitoring, which warrants a larger, longitudinal study to validate these results.

Advancing Optical Coherence Tomography Diagnostic Capabilities: Machine Learning Approaches to Detect Autoimmune Inflammatory Diseases.

In many parts of the world including India, the prevalence of autoimmune inflammatory diseases such as neuromyelitis optica spectrum disorders (NMOSD), myelin oligodendrocyte glycoprotein antibody-associated disease (MOGAD), and multiple sclerosis (MS) is rising. A diagnosis is often delayed due to insufficient diagnostic tools. Machine learning (ML) models have accurately differentiated eyes of patients with MS from those of healthy controls (HCs) using optical coherence tomography (OCT)-based retinal images. Examining OCT characteristics may allow for early differentiation of these conditions. The objective of this study was to determine feasibility of ML analyses to distinguish between patients with different autoimmune inflammatory diseases, other ocular diseases, and HCs based on OCT measurements of the peripapillary retinal nerve fiber layer (pRNFL), ganglion cell-inner plexiform layer (GCIPL), and inner nuclear layers (INLs). Eyes of people with MS (n = 99 patients), NMOSD (n = 40), MOGAD (n = 74), other ocular diseases (OTHER, n = 16), and HCs (n = 54) from the Mangalore Demyelinating Disease Registry were included. Support vector machine (SVM) classification models incorporating age, pRNFL, GCIPL, and INL were performed. Data were split into training (70%) and testing (30%) data and accounted for within-patient correlations. Cross-validation was used in training to choose the best parameters for the SVM model. Accuracy and area under receiver operating characteristic curves (AUROCs) were used to assess model performance. The SVM models distinguished between eyes of patients with each condition (i.e., MOGAD vs NMOSD, NMOSD vs HC, MS vs OTHER, etc) with strong discriminatory power demonstrated from the AUROCs for these comparisons ranging from 0.81 to 1.00. These models also performed with moderate to high accuracy, ranging from 0.66 to 0.81, with the exception of the MS vs NMOSD comparison, which had an accuracy of 0.53. ML models are useful for distinguishing between autoimmune inflammatory diseases and for distinguishing these from HCs and other ocular diseases based on OCT measures. This study lays the groundwork for future deep learning studies that use analyses of raw OCT images for identifying eyes of patients with such disorders and other etiologies of optic neuropathy.

Ensemble learning to predict short birth interval among reproductive-age women in Ethiopia: evidence from EDHS 2016–2019

BackgroundA birth interval of less than 33 months was considered short, and in low- income countries like Ethiopia, a short birth interval is the primary cause of approximately 822 maternal deaths every day. Due to that this study aimed to predict short birth interval and associated factors among women (15–49) in Ethiopia using ensemble learning algorithms.MethodsA secondary data analysis of Ethiopian demographic health servey from 2016 to 2019 was performed. a total of weighted sample of 12,573 women in the reproductive age group was included in this study. Data have been extracted and processed with Stata version 17. The dataset was then imported into a Jupyter notebook for further detailed analysis and visualization. An ensemble Machin learning algorithm using different classification models were implemented. All analysis and calculation were performed using Python 3 programming language in Jupyter Notebook using imblearn, sklearn, and xgboost pakages.ResultsRandom forest demonstrated the best performance with an accuracy 97.84%, recall of 99.70%, F1-score of 97.81%, 98.95% precision on test data and AUC (98%). Region, residency, age of women, sex of child, respondent education, distance health facility, husband education and religion were top predicting factors of short birth interval among women in Ethiopia.ConclusionRandom forest was best predictive models with improved performance. "The most significant features that contribute to the accuracy of the top-performing models, notably the Random Forest should be highlighted because they outperformed the other model in the analysis.In general, ensemble learning algorithms can accurately predict short birth interval status, making them potentially useful as decision-support tools for the pertinent stakeholders.

A short investigation of the effect of the selection of human brain atlases on the performance of ASD's classification models

This study investigated the impact of brain atlas selection on the classification accuracy of Autism Spectrum Disorder (ASD) models using functional Magnetic Resonance Imaging (fMRI) data. Brain atlases, such as AAL, CC200, Harvard-Oxford, and Yeo 7/17, are used to define regions of interest (ROIs) for fMRI analysis and play a crucial role in enabling researchers to study connectivity patterns and neural dynamics in ASD patients. Through a systematic review, we examined the performance of different atlases in various machine-learning and deep-learning frameworks for ASD classification. The results reveal that atlas selection significantly affects classification accuracy, with denser atlases, such as CC400, providing higher granularity, whereas coarser atlases such as AAL, offer computational efficiency. Furthermore, we discuss the dynamics of combining multiple atlases to enhance feature extraction and explore the implications of atlas selection across diverse datasets. Our findings emphasize the need for standardized approaches to atlas selection and highlight future research directions, including the integration of novel atlases, advanced data augmentation techniques, and end-to-end deep-learning models. This study provides valuable insights into optimizing fMRI-based ASD diagnosis and underscores the importance of interpreting atlas-specific features for an improved understanding of brain connectivity in ASD.

A Review of Tree‐Based Methods for Analyzing Survey Data

ABSTRACTRecent advances in data complexity and availability present both challenges and opportunities for automated data exploration. Tree‐based methods, known for their interpretability, are widely used for building regression and classification models. However, they often lag behind the best supervised learning approaches in terms of prediction accuracy. To address this limitation, ensemble methods, such as random forests, combine multiple trees to improve prediction accuracy, though at the cost of interpretability. While tree‐based methods have seen extensive use in various fields, their application in the context of complex survey data has been relatively limited. This article provides an overview of the state‐of‐the‐art tree‐based approaches for analyzing complex survey data. It distinguishes methods explicitly designed for survey contexts from those adapted from other domains. The discussion covers applications in model‐assisted approaches, disclosure limitation, and small area estimation, as well as other recent methodological developments tailored to survey data. Additionally, the article explores aggregated tree models that sacrifice interpretability for improved prediction accuracy. These models, such as Bagging, Random Forests, and Boosting, are explained, along with the concept of out‐of‐bag error for model evaluation. Finally, this article provides the history and development of tree models, from their origins in regression trees to more recent Bayesian approaches, and aggregated tree models. This overview sheds light on the potential utility of tree‐based methods in survey methodology and provides insights into future research directions in this evolving field.

Lower Limb Exoskeletons, Application‐Centric Classifications: A Review

ABSTRACTExoskeletons are a relatively new field of technology used as assistive devices. Since the first development of the exoskeleton in the late 19th century, this field has seen significant advancements, as it has potential in numerous applications. Although this field is rapidly growing, it remains unable to be fully implemented in various applications. It is due to the limitations in the existing technology used to develop the exoskeleton, the lack of standardization, and the rapid growth in the exoskeleton field. To address the existing challenge, this review paper attempts to develop a classification model of lower limb exoskeletons (LLEs), which researchers and engineers can use to understand the relation between various application scenarios and different design aspects while developing an LLE. The classification model has been divided into two levels: application and design aspects of an exoskeleton. In this paper, 53 LLEs have been analyzed and classified into a flexible model, allowing future exoskeletons also to be added. It has also been discussed why each exoskeleton is classified into various categories.

A Deep-Learning Model for Multi-class Audio Classification of Vocal Fold Pathologies in Office Stroboscopy.

To develop and validate a deep-learning classifier trained on voice data extracted from videolaryngostroboscopy recordings, differentiating between three different vocal fold (VF) states: healthy (HVF), unilateral paralysis (UVFP), and VF lesions, including benign and malignant pathologies. Patients with UVFP (n = 105), VF lesions (n = 63), and HVF (n = 41) were retrospectively identified. Voice samples were extracted from stroboscopic videos (Pentax Laryngeal Strobe Model 9400), including sustained /i/ phonation, pitch glide, and /i/ sniff task. Extracted audio files were converted into Mel-spectrograms. Voice samples were independently divided into training (80%), validation (10%), and test (10%) by patient. Pretrained ResNet18 models were trained to classify (1) HVF and pathological VF (lesions and UVFP), and (2) HVF, UVFP, and VF lesions. Both classifiers were further validated on an external dataset consisting of 12 UVFP, 13 VF lesions, and 15 HVF patients. Model performances were evaluated by accuracy and F1-score. When evaluated on a hold-out test set, the binary classifier demonstrated stronger performance compared to the multi-class classifier (accuracy 83% vs. 40%; F1-score 0.90 vs. 0.36). When evaluated on an external dataset, the binary classifier achieved an accuracy of 63% and F1-score of 0.48, compared to 35% and 0.25 for the multi-class classifier. Deep-learning classifiers differentiating HVF, UVFP, and VF lesions were developed using voice data from stroboscopic videos. Although healthy and pathological voice were differentiated with moderate accuracy, multi-class classification lowered model performance. The model performed poorly on an external dataset. Voice captured in stroboscopic videos may have limited diagnostic value, though further studies are needed. 4 Laryngoscope, 2025.

Integration of CNN Models and Machine Learning Methods in Credit Score Classification: 2D Image Transformation and Feature Extraction

Abstract The problem of accurately classifying credit scores is critical for financial institutions to assess individual creditworthiness and effectively manage credit risk. Traditional methods often face limitations when processing large datasets, resulting in lower accuracy and longer processing time. To address this issue, this paper proposes a novel approach to credit score classification by integrating convolutional neural networks (CNN) with machine learning methods. First, a 1D dataset of sequential text data is transformed into 2D greyscale images to use 2D CNN models for feature extraction and classification. Six CNN architectures—DenseNet201, GoogLeNet, MobileNetV2, ResNet18, ShuffleNet, and SqueezeNet—are implemented, and the features in the last layer (1000 features) of each CNN are classified using the softmax method. To further improve the performance, the two best CNN models were selected, and a new fully connected layer (NewFC) was added. A class-based feature set [3 × 31,695] representing three credit score types (good, poor, and standard) was extracted from each model and merged into a feature set [6 × 31,695]. This combined feature set was then reclassified using KNN, LDA, Naive Bayes, and SVM algorithms. The performance of both CNN and machine learning methods was evaluated using accuracy, precision, sensitivity, specificity, and F-score metrics. To optimize classification performance and reduce computational cost, the RelieF algorithm was used to select the best 5 out of 6 features. Compared to using all 6 features, significant improvements in accuracy and efficiency were observed, demonstrating the effectiveness of the proposed method in credit score classification.

Deep Learning-Based Point Cloud Classification of Obstacles for Intelligent Vehicles

Intelligent driving research has focused much attention on point cloud obstacles since they are a class of high-dimensional data that can adequately depict the shape and placement of obstacles, unlike picture data. Currently, deep learning technology is primarily employed for vehicle autonomy point cloud obstacle classification tasks. These techniques typically struggle with low classification accuracy, processing efficiency, and model stability. To tackle the abovementioned issues, this paper suggests a novel random forest algorithm that integrates the out-of-bag error theory and can consistently and accurately evaluate the influence of point cloud properties. Then, building on the novel algorithm, this paper suggests a modified PointNet network that incorporates the effects of both global and local features on the classification task, therefore increasing the conventional network’s classification accuracy. To assess the effectiveness of this novel approach in the experimental portion, we set up an evaluation system based on the metrics for average accuracy, overall accuracy, and a confusion matrix. According to the simulation results, the overall accuracy of the proposed network in terms of classification accuracy is 94.4% and the average accuracy is 84.9%, which are then compared to the prototype PointNet and its variants. The classification accuracies for the four types of obstacles are 97.6%, 63.6%, 92.5%, and 86.1%. In addition, the proposed method is effective at improving both the computational complexity and stability of the network.

Development of an artificial intelligence-based application for the diagnosis of sarcopenia: a retrospective cohort study using the health examination dataset

BackgroundMedical imaging techniques for diagnosing sarcopenia have been extensively investigated. Studies have proposed using the T-score and patient information as key diagnostic factors. However, these techniques have either been time-consuming or have required separate calculation processes after collecting each parameter. To address this gap, we propose an artificial intelligence (AI)-based web application that automates the collection of data, classification of the lumbar spine 3 (L3) slices, segmentation of the subcutaneous fat, visceral fat, and muscle areas in the classified L3 slices, and quantitative analysis of the segmented areas.MethodsWe developed an automated lumbar spine slice classification model using the CNN (EfficientNetV2) algorithm and an automated domain segmentation model to identify the subcutaneous fat, visceral fat, and muscle areas using the U-NET algorithm. These models were used to identify L3 slices from abdominal computed tomography images and divide the images into the three-segmented domains for sarcopenia diagnosis. Additionally, we developed an algorithm for the calculation of T-Score calculated as (measurement value-Young adult mean)/(Young adult SD) using the Aggregation Pipeline by MongoDB, with the mean and standard deviation for skeletal muscle area (SMA), SMA/height2, SMA/weight, and SMA/body mass index (BMI) for both sexes and different age groups.ResultsThe proposed system demonstrated high accuracy and precision, with an overall accuracy of 97.5% in classifying L3 slices and a segmentation accuracy of 92% for muscle, subcutaneous fat, and visceral fat areas. The T-Score-based analysis provided reliable diagnostic thresholds for sarcopenia, facilitating consistent and accurate assessments. Our diagnostic cutoff points for each index were as follows: SMA (-1.0: 152.55, -2.0: 125.89), SMA/height² (-1.0: 38.84, -2.0: 14.50), SMA/weight (-1.0: 2.14, -2.0: 1.89), and SMA/BMI (-1.0: 6.10, -2.0: 5.18) for men; SMA (-1.0: 96.08, -2.0: 76.96), SMA/height² (-1.0: 37.20, -2.0: 29.36), SMA/weight (-1.0: 1.80, -2.0: 1.61), and SMA/BMI (-1.0: 4.56, -2.0: 4.01) for women. SMA/BMI best reflected the loss of muscle mass in healthy populations by age, showing a more remarkable decrease in muscle mass in men than in women. The values for men gradually decreased after their 20s, and that for women gradually decreased after their 40s, which progressed to a more dramatic decline in the 70s for both sexes.ConclusionThis AI-based web application addresses the limitations of previous diagnostic techniques by automatically analyzing medical images for the classification, segmentation, and calculation of T-scores. The study findings provide a more reliable and accurate diagnostic technique for sarcopenia that can consequently impact patient treatment and outcomes.

Identification and Localization Study of Grounding System Defects in Cross-Bonded Cables

Cross-bonded cables improve transmission efficiency by optimizing the grounding method. However, due to the complexity of their grounding system, they are prone to multiple types of defects, making defect state identification more challenging. Additionally, accurately locating sheath damage defects becomes more difficult in cases of high transition resistance. To address these issues, this paper constructs a distributed parameter circuit model for cross-bonded cables and proposes a particle swarm optimization support vector machine (PSO-SVM) defect classification model based on the sheath voltage and current phase angle and amplitude characteristics. This model effectively classifies 25 types of grounding system states. Furthermore, for two types of defects—open joints and sheath damage short circuits—this paper proposes an accurate segment-based location method based on fault impedance characteristics, using zero-crossing problems to achieve efficient localization. The results show that the distributed parameter circuit model for cross-bonded cables is feasible for simulating electrical quantities, as confirmed by both simulation and real-world applications. The defect classification model achieves an accuracy of over 97%. Under low transition resistance, the defect localization accuracy exceeds 95.4%, and the localization performance is significantly improved under high transition resistance. Additionally, the defect localization method is more sensitive to variations in cable segment length and grounding resistance impedance but less affected by fluctuations in core voltage and current.

SPOCK 2.0: Updates to the FeatureClassifier in the Stability of Planetary Orbital Configurations Klassifier

Abstract The Stability of Planetary Orbital Configurations Klassifier (SPOCK) package collects machine learning models for predicting the stability and collisional evolution of compact planetary systems. In this paper we explore improvements to SPOCK’s binary stability classifier (FeatureClassifier), which predicts orbital stability by collecting data over a short N-body integration of a system. We find that by using a system-specific timescale (rather than a fixed 104 orbits) for the integration, and by using this timescale as an additional feature, we modestly improve the model’s AUC metric from 0.943 to 0.950 (AUC = 1 for a perfect model). We additionally discovered that ≈10% of N-body integrations in SPOCK’s original training data set were duplicated by accident, and that &lt;1% were misclassified as stable when they in fact led to ejections. We provide a cleaned data set of 100,000+ unique integrations, release a newly trained stability classification model, and make minor updates to the API.

Artificial intelligence-based cardiac transthyretin amyloidosis detection and scoring in scintigraphy imaging: multi-tracer, multi-scanner, and multi-center development and evaluation study.

Providing tools for comprehensively evaluating scintigraphy images could enhance transthyretin amyloid cardiomyopathy (ATTR-CM) diagnosis. This study aims to automatically detect and score ATTR-CM in total body scintigraphy images using deep learning on multi-tracer, multi-scanner, and multi-center datasets. In the current study, we employed six datasets (from 12 cameras) for various tasks and purposes. Dataset #1 (93 patients, 99mTc-MDP) was used to develop the 2D-planar segmentation and localization models. Dataset #2 (216 patients, 99mTc-DPD) was used for the detection (grade 0 vs. grades 1, 2, and 3) and scoring (0 and 1 vs. grades 2 and 3) of ATTR-CM. Datasets #3 (41 patients, 99mTc-HDP), #4 (53 patients, 99mTc-PYP), and #5 (129 patients, 99mTc-DPD) were used as external centers. ATTR-CM detection and scouring were performed by two physicians in each center. Moreover, Dataset #6 consisting of 3215 patients without labels, was employed for retrospective model performance evaluation. Different regions of interest were cropped and fed into the classification model for the detection and scoring of ATTR-CM. Ensembling was performed on the outputs of different models to improve their performance. Model performance was measured by classification accuracy, sensitivity, specificity, and AUC. Grad-CAM and saliency maps were generated to explain the models' decision-making process. In the internal test set, all models for detection and scoring achieved an AUC of more than 0.95 and an F1 score of more than 0.90. For detection in the external dataset, AUCs of 0.93, 0.95, and 1 were achieved for datasets 3, 4, and 5, respectively. For the scoring task, AUCs of 0.95, 0.83, and 0.96 were achieved for these datasets, respectively. In dataset #6, we found ten cases flagged as ATTR-CM by the network. Out of these, four cases were confirmed by a nuclear medicine specialist as possibly having ATTR-CM. GradCam and saliency maps showed that the deep-learning models focused on clinically relevant cardiac areas. In the current study, we developed and evaluated a fully automated pipeline to detect and score ATTR-CM using large multi-tracer, multi-scanner, and multi-center datasets, achieving high performance on total body images. This fully automated pipeline could lead to more timely and accurate diagnoses, ultimately improving patient outcomes.

Realistic Data Delays and Alternative Inactivity Definitions in Telecom Churn: Investigating Concept Drift Using a Sliding-Window Approach

Predicting customer churn is essential for telecommunications companies to maintain profitability. However, training models on historical models leads to performance degradation when they are applied to future conditions—a phenomenon known as concept drift. We employ a sliding-window approach that separates the training and testing time windows, creating a future-based “true test”. Using unique real data, we show that a CatBoost classifier model trained on older data can remain relevant when new, unseen intervals are used. A key innovation of our work is the use of 40-day “partial churn” labels; a model trained on these labels accurately predicts 90-day churn by simply adjusting the decision threshold. Out of the six modeled scenarios, in the main realistic scenario, CatBoost retained an accuracy above 0.798 and an F1 of near 0.704, reflecting its robustness even under real-world delays and potential drift. Overall, our findings emphasize that models do not necessarily “expire” with time; rather, their performance varies according to when they are tested. This research underscores the importance of a truly future-based evaluation (instead of artificial splits) and offers practical guidance for earlier churn detection when facing real-world data delays.

Prediction of hemolytic peptides and their hemolytic concentration.

Peptide-based drugs often fail in clinical trials due to their toxicity or hemolytic activity against red blood cells (RBCs). Existing methods predict hemolytic peptides but not the concentration (HC50) required to lyse 50% of RBCs. This study develops classification and regression models to identify and quantify hemolytic activity. These models train on 1926 peptides with experimentally determined HC50 against mammalian RBCs. Analysis indicates that hydrophobic and positively charged residues were associated with higher hemolytic activity. Among classification models, including machine learning (ML), quantum ML, and protein language models, a hybrid model combining random forest (RF) and a motif-based approach achieves the highest area under the receiver operating characteristic curve (AUROC) of 0.921. Regression models achieve a Pearson correlation coefficient (R) of 0.739 and a coefficient of determination (R²) of 0.543. These models outperform existing methods and are implemented in HemoPI2, a web-based platform and standalone software for designing peptides with desired HC50 values ( http://webs.iiitd.edu.in/raghava/hemopi2/ ).

Deep learning interpretability analysis for carbon star identification in Gaia DR3

A large fraction of asymptotic giant branch (AGB) stars develop carbon-rich atmospheres during their evolution. Based on their color and luminosity, these carbon stars can easily be distinguished from many other kinds of stars. However, numerous G, K, and M giants also occupy the same region as carbon stars on the HR diagram. Despite this fact, their spectra exhibit differences, especially in the prominent CN molecular bands. We aim to distinguish carbon stars from other kinds of stars using Gaia's XP spectra while providing attributional interpretations of key features that are necessary for identification and even discovering new key spectral features. We propose a classification model named "GaiaNet," an improved one-dimensional convolutional neural network specifically designed for handling Gaia's XP spectra. We utilized SHapley Additive exPlanations (SHAP), an approach for interpretability based on game theoretic, to determine SHAP values for each feature in a spectrum, enabling us to explain the output of the GaiaNet model and provide further meaningful analysis. Compared to four traditional machine learning methods, the GaiaNet model exhibits an average classification accuracy improvement of approximately 0.3% on the validation set, with the highest accuracy reaching 100%. Utilizing the SHAP model, we present a clear spectroscopic heatmap highlighting molecular band absorption features primarily distributed around CN_ and CN_ and we summarize five key feature regions for carbon star identification. Upon applying the trained classification model to the CSTAR sample with Gaia "xp_sampled_mean" spectra, we obtained 451 new candidate carbon stars as a by-product. Our algorithm is capable of discerning subtle feature differences from low-resolution spectra of Gaia, thereby assisting us in effectively identifying carbon stars with typically higher temperatures and weaker CN features while providing compelling attributive explanations. The interpretability analysis of deep learning holds significant potential in spectral identification.