Classification Approach Research Articles

Multiclass hate speech detection with an aggregated dataset

Abstract Detecting and removing hate speech content in a timely manner remains a challenge for social media platforms. Automated techniques such as deep learning models offer solutions which can keep up with the volume and velocity of user content production. Research in this area has mainly focused on either binary classification or on classifying tweets into generalised categories such as hateful, offensive, or neither. Less attention has been given to multiclass classification of online hate speech into the type of hate or group at which it is directed. By aggregating and re-annotating several relevant hate speech datasets, this study presents a dataset and evaluates several models for classifying tweets into the categories ethnicity, gender, religion, sexuality, and non-hate. We evaluate the dataset by training several models: logistic regression, LSTM, BERT, and GPT-2. For the LSTM model, we assess a range of NLP features using a multi-classification LSTM model, and conclude that the highest performing feature combination consists of word $n$ -grams, character $n$ -grams, and dependency tuples. We show that while more recent larger models can achieve a slightly higher performance, increased model complexity alone is not sufficient to achieve significantly improved models. We also compare this approach with a binary classification approach and evaluate the effect of dataset size on model performance.

An Ensemble Voting classification Approach for Software defects prediction

An Ensemble Voting classification Approach for Software defects prediction

Morphological Classification of Galaxies Through Structural and Star Formation Parameters Using Machine Learning

Abstract We employ the XGBoost machine learning (ML) method for the morphological classification of galaxies into two (early-type, late-type) and five (E, S0–S0a, Sa–Sb, Sbc–Scd, Sd–Irr) classes, using a combination of non-parametric (C, A, S, AS, Gini, M20, c5090), parametric (Sérsic index, n), geometric (axial ratio, BA), global colour (g − i, u − r, u − i), colour gradient (Δ(g − i)), and asymmetry gradient (ΔA9050) information, all estimated for a local galaxy sample (z < 0.15) compiled from the Sloan Digital Sky Survey (SDSS) imaging data. We train the XGBoost model and evaluate its performance through multiple standard metrics. Our findings reveal better performance when utilizing all fourteen parameters, achieving accuracies of 88% and 65% for the two-class and five-class classification tasks, respectively. In addition, we investigate a hierarchical classification approach for the five-class scenario, combining three XGBoost classifiers. We observe comparable performance to the “direct” five-class classification, with discrepancies of only up to 3%. Using SHAP (an advanced interpretation tool), we analyse how galaxy parameters impact the model’s classifications, providing valuable insights into the influence of these features on classification outcomes. Finally, we compare our results with previous studies and find them consistently aligned.

Deep learning empowered sensor fusion boosts infant movement classification

BackgroundTo assess the integrity of the developing nervous system, the Prechtl general movement assessment (GMA) is recognized for its clinical value in diagnosing neurological impairments in early infancy. GMA has been increasingly augmented through machine learning approaches intending to scale-up its application, circumvent costs in the training of human assessors and further standardize classification of spontaneous motor patterns. Available deep learning tools, all of which are based on single sensor modalities, are however still considerably inferior to that of well-trained human assessors. These approaches are hardly comparable as all models are designed, trained and evaluated on proprietary/silo-data sets.MethodsWith this study we propose a sensor fusion approach for assessing fidgety movements (FMs). FMs were recorded from 51 typically developing participants. We compared three different sensor modalities (pressure, inertial, and visual sensors). Various combinations and two sensor fusion approaches (late and early fusion) for infant movement classification were tested to evaluate whether a multi-sensor system outperforms single modality assessments. Convolutional neural network (CNN) architectures were used to classify movement patterns.ResultsThe performance of the three-sensor fusion (classification accuracy of 94.5%) is significantly higher than that of any single modality evaluated.ConclusionsWe show that the sensor fusion approach is a promising avenue for automated classification of infant motor patterns. The development of a robust sensor fusion system may significantly enhance AI-based early recognition of neurofunctions, ultimately facilitating automated early detection of neurodevelopmental conditions.

Assessing the impacts of forest cover change on carbon stock and soil moisture dynamics using geospatial techniques: a case study of Nensebo forest, Southern Ethiopia.

Assessing the impacts of forest cover change on carbon stock and soil moisture dynamics is critical for understanding environmental degradation and guiding sustainable land management. This study evaluates the effects of forest cover change on carbon stock and soil moisture dynamics in Nensebo Forest from 1993 to 2023 using geospatial techniques. Landsat imagery including TM (1993), ETM + (2009), and OLI/TIRS (2023) were used. Land use land cover (LULC) classification was performed using supervised classification approach with maximum likelihood algorithm. The study reveals that agricultural land expanded significantly from 143.2 km2 (21.8%) to 230.0 km2 (35.1%), along with increases in bare land and grassland. Moreover, result reveals a substantial decline in vegetation health (forest), with the Normalized Difference Vegetation Index (NDVI) decreasing from 0.78 in 1993 to 0.55 in 2023, indicating anincrease ofdeforestation and land degradation. The soil moisture index (SMI) shows a reduction in soil moisture retention from 0.51 to 0.39, reflecting the adverse impact of forest loss on soil moisture dynamics. Carbon stock analysis reveals a dramatic decrease in forest area from 489.5 to 220.5 km2, leading to increased carbon dioxide emissions from forest loss, which rose from 293.03 tons to 1550.34 tons. The study underscores the urgent need for integrated land management strategies that balance agricultural development with ecological conservation. Therefore, to mitigate climate change and maintain ecological stability, the study recommends implementing sustainable land use practices, promoting reforestation, and developing environmental relatedpolicies to preserve the remaining forest cover while supporting agricultural productivity.

Early detection of canine hemangiosarcoma via cfDNA fragmentation and copy number alterations in liquid biopsies using machine learning

Hemangiosarcoma is a highly malignant tumor commonly affecting canines, originating from endothelial cells that line blood vessels, underscoring the importance of early detection. This canine cancer is analogous to human angiosarcoma, and the development of liquid biopsies leveraging cell-free DNA (cfDNA) represents a promising step forward in early cancer diagnosis. In this study, we utilized Whole Genome Sequencing (WGS) to analyze fragment sizes and copy number alterations (CNAs) in cfDNA from 21 hemangiosarcoma-affected and 36 healthy dogs, aiming to enhance early cancer detection accuracy through machine learning models. Our findings reveal that similar to trends in human oncology, hemangiosarcoma samples exhibited shorter DNA fragment sizes compared to healthy controls, with a notable leftward shift in the primary peak. Interestingly, canine hemangiosarcoma DNA fragment sizes demonstrated eight distinct periodic patterns diverging from those typically observed in human angiosarcoma. Additionally, we identified seven novel genomic gains and nine losses in the hemangiosarcoma samples. Applying machine learning to the cfDNA fragment size distribution, we achieved an impressive average Area Under the Curve (AUC) of 0.93 in 10-fold cross-validation, underscoring the potential of this approach for precise early-stage cancer classification. This study confirms distinctive cfDNA fragment size and CNA patterns in hemangiosarcoma-affected vs. healthy dogs and demonstrates the promise of these biomarkers in canine cancer screening, early detection, and monitoring via liquid biopsies. These findings establish a foundation for broader research on cfDNA analysis in various canine cancers, integrating methodologies from human oncology to enhance early detection and diagnostic precision in veterinary medicine.

Music Emotion Classification System Based on Quantum Feature Extraction and Its Effectiveness Evaluation

Strong emotions can be expressed and evoked via music. However, accurately recognizing the emotions in music using computational models is extremely challenging. When the music portions convey various complex emotions, the problem’s difficulty can rise significantly. In this work, a systematic strategy that blends multiple cutting-edge techniques is used to focus on the emotional classification of music. The Kaggle-sourced music emotion dataset provides a wide range of audio samples illustrating various emotional expressions in music. Spectral subtraction as a noise reduction approach is used to improve the quality of the analysis. This efficiently reduced unwanted background noise and improved the audio signals’ clarity. A Quantum Convolutional Neural Network (QCNN) is used for feature extraction, taking advantage of its special capacity to extract complex patterns and characteristics from the music data that are essential for emotion recognition. To provide accurate predictions, the training procedure is optimized and finally utilized Flexible Runge Kutta Optimized Multilayer Perceptrons (FRKO-MLP) for emotion classification. The goal of this combination of advanced classification approaches, quantum feature extraction and noise reduction is to increase the robustness and accuracy of music emotion identification systems. Based on the chosen datasets, the results of the experimental study demonstrate that the suggested model greatly increases the efficiency and accuracy of emotion classification in music. The proposed FRKO-MLP attains 93.21% accuracy, 90.52% precision, 87.34% sensitivity and 89.72% F1-score. These findings show that the model is far better at identifying emotions than traditional methods.

Capacity-building interventions for health extension workers in Ethiopia: A scoping review.

Capacity-building interventions for health extension workers (HEWs) are key to providing quality health services to the community. Since Ethiopia's Health Extension Program was established, several types of capacity-building interventions have been developed to build HEW competencies. However, no comprehensive study has mapped the types of capacity-building interventions being used or the competencies targeted. To (1) identify and characterize evidence on capacity-building interventions for Ethiopian HEWs, including the competencies measured; (2) clarify evidence gaps in this area; and (3) explore how successful the interventions have been to inform the design of health extension programs and further research. We used keywords (health extension workers, capacity building, competencies) and related terminologies to search PubMed, Scopus, and Embase for published studies on capacity-building interventions for Ethiopian HEWs, and Google Scholar for unpublished studies and reports. Our search was limited to studies and reports published in English from 2003 to present. We used the JBI scoping review methodology to conduct this scoping review in a stepwise approach and a categorization approach to synthesize the evidence. Our search strategy identified 20 articles, all published except for one program report. The most common capacity-building intervention designed for HEWs was training, followed by supportive supervision, performance review and clinical mentoring meetings, and equipment supply; the most salient competency domains investigated were knowledge and skills. The interventions significantly improved immediate outcomes (knowledge, skills, attitude change among HEWs) and intermediate outcomes, such as increased service utilization and health-seeking behavior among community members. Only one study assessed whether capacity-building interventions improved inter- and intra-personal domains of capacity/competency. Capacity-building interventions for Ethiopian HEWs were found to be effective, but they mainly focused on improving technical competencies, such as knowledge and skills. Little attention has been paid to other competency domains, including motivation, leadership, and communication. Thus, future research could focus on a comprehensive set of capacity-building initiatives that addresses motivation, job satisfaction, communication, commitment, and resource allocation.

Wood Waste Valorization and Classification Approaches: A systematic review

Wood Waste Valorization and Classification Approaches: A systematic review

Molecular Dynamics of Peptide Sequencing through MoS2 Solid-State Nanopores for Binary Encoding Applications.

Biological peptides have emerged as promising candidates for data storage applications due to their versatility and programmability. Recent advances in peptide synthesis and sequencing technologies have enabled the development of peptide-based data storage systems for realizing novel information storage technologies with enhanced capacity, durability, and data access speeds. In this study, we performed coarse-grained peptide sequencing of 12 distinct sequences through single-layer MoS2 solid-state nanopores (SSNs) using molecular dynamics (MD). Peptide sequences were composed of 1 positively charged, 1 negatively charged, and 4 neutral amino acids, with the position of amino acids in the sequence being shuffled to generate all possible configurations. From MD, the goal was to evaluate the efficiency of these peptide sequences to encode binary information based on ionic current traces monitored during their passage through the SSNs. Classification approaches using LightGBM were trained and tested to analyze different sequence factors such as the position of amino acids or the spacing between charged amino acids in the sequences. Our findings reveal the presence of two distinct groups of sequences determined by the relative position of the positively charged amino acid compared to the negatively charged amino acid. Furthermore, we observe a strong correlation between discrimination accuracy and the separation in the sequence between charged amino acids, depending on the number of adjacent neutral amino acids between them. Finally, MD allowed us to establish the nonlinear relationship between amino acid positions inside the pore (called sequence motifs) and fluctuations in ionic current traces to discriminate false positives and to enable effective training of machine learning classification algorithms. These very promising results emphasized the best approaches to design peptide sequences as building blocks for molecular data storage. Finally, this study highlights the potential of the proposed approach for designing peptide sequence combinations that could help the development of efficient, scalable, and reliable molecular data storage solutions, with future research focused on encoding longer binary chains to enhance storage capacity and support the goal of stable, energy-free biological systems.

Redefining the severity of orofacial tissue damage caused by noma: a novel classification approach.

Noma is a severe orofacial disease with high mortality and morbidity. Although severity scales exist, they fail to fully capture the extent of damage caused by the disease. This study analysed 404 photos of 260 noma cases from Facing Africa (n=228) and Project Harar (n=32) to create a new severity classification system. Photos were displayed on a large screen for better visualization. Tissue damage was assessed in terms of width, depth and surface area using the American Burn Association's grading system. A mathematical formula was also developed to quantify the extent of noma-induced orofacial damage. Fifteen orofacial anatomical units vulnerable to noma were identified, leading to the development of a new severity classification system that enhances the assessment of noma. This system evaluates tissue damage qualitatively as minor, moderate or major and quantitatively using a purpose-built mathematical formula. Current classification systems inadequately capture noma severity due to vague definitions and limited consideration of the affected anatomical units. The new approach addresses these gaps by considering all tissue damage, including dentition, and allows for precise qualitative and quantitative assessment of disease severity. Continued research and validation are essential to refine this method, improving its clinical and research applications and enhancing outcomes for noma patients worldwide.

Diagnosing Helicobacter pylori using autoencoders and limited annotations through anomalous staining patterns in IHC whole slide images.

This work addresses the detection of Helicobacter pylori (H. pylori) in histological images with immunohistochemical staining. This analysis is a time-demanding task, currently done by an expert pathologist that visually inspects the samples. Given the effort required to localize the pathogen in images, a limited number of annotations might be available in an initial setting. Our goal is to design an approach that, using a limited set of annotations, is capable of obtaining results good enough to be used as a support tool. We propose to use autoencoders to learn the latent patterns of healthy patches and formulate a specific measure of the reconstruction error of the image in HSV space. ROC analysis is used to set the optimal threshold of this measure and the percentage of positive patches in a sample that determines the presence of H. pylori. Our method has been tested on an own database of 245 whole slide images (WSI) having 117 cases without H. pylori and different density of the bacteria in the remaining ones. The database has 1211 annotated patches, with only 163 positive patches. This dataset of positive annotations was used to train a baseline thresholding and an SVM using the features of a pre-trained RedNet-18 and ViT models. A 10-fold cross-validation shows that our method has better performance with 91% accuracy, 86% sensitivity, 96% specificity and 0.97 AUC in the diagnosis of H. pylori . Unlike classification approaches, our shallow autoencoder with threshold adaptation for the detection of anomalous staining is able to achieve competitive results with a limited set of annotated data. This initial approach is good enough to be used as a guide for fast annotation of infected patches.

RFMVDA: An Enhanced Deep Learning Approach for Customer Behavior Classification in E-Commerce Environments

RFMVDA: An Enhanced Deep Learning Approach for Customer Behavior Classification in E-Commerce Environments

GNN4HT: A Two-Stage GNN-Based Approach for Hardware Trojan Multifunctional Classification

GNN4HT: A Two-Stage GNN-Based Approach for Hardware Trojan Multifunctional Classification

A deep learning-based approach for two-phase flow pattern classification using void fraction time series analysis

A deep learning-based approach for two-phase flow pattern classification using void fraction time series analysis

Neutrosophic Sets in Big Data Analytics: A Novel Approach for Feature Selection and Classification

Big Data Analytics are said to help in transforming huge amounts of raw data towards valuable information that can be used, but there are formidable challenges in feature selection and classification due to the complexity and high dimensionality of the data. Traditional methods are usually too weak to handle the built-in uncertainty, imprecision, and inconsistency within big data and they often fail to perform well. This paper aims to induce the new methodology on these problems using the sets of neutrosophic in dealing with more flexible and nuanced data analysis. The key contributions to the current approach proposed are threefold. First, generalization of the classical set through extension of the notions of truth, indeterminacy, and falsity by allowing representations of uncertainty in data. The second combines a powerful process for selecting features based upon neutrosophic set theory that is optimal by genetic algorithms and advances a step further by applying these features in training and validating the classification models across a set of different domains. Therefore, the major aim from this study is to increase accuracy and reliability in feature selection and classification in big data analytics. This methodology has been implemented and tested over datasets of the following types: healthcare, finance, social media, and more. Results have proved great improvement against conventional performance metrics, for example, the classification accuracy with an SVM classifier over the Cleveland Heart Disease dataset increases from 83.5% to 87.2%, and of a Random Forest classifier over a financial dataset from 76.4% to 81.9%. For instance, the accuracy of social media sentiment analysis changed to 82.7% from 78.3%. All these findings establish that the neutrosophic set-based method holds good advantages in addressing the limitations of classical alternatives. The proposed approach of neutrosophism, through an explicit model, enhances performances in classifications and, at the same time, augments overall robustness and reliability in big data analytic. The importance of this study lies in establishing the groundwork for further research and practical applications, thus indicating possible further development in this field.

Machine Learning Approaches for Binary Classification of Sorghum (Sorghum bicolor L.) Seeds from Image Color Features

Machine Learning Approaches for Binary Classification of Sorghum (Sorghum bicolor L.) Seeds from Image Color Features

A supervised machine learning approach for structural overload classification in railway bridges using weigh-in-motion data

A supervised machine learning approach for structural overload classification in railway bridges using weigh-in-motion data

Knowledge distillation approach for skin cancer classification on lightweight deep learning model

AbstractOver the past decade, there has been a global increase in the incidence of skin cancers. Skin cancer has serious consequences if left untreated, potentially leading to more advanced cancer stages. In recent years, deep learning based convolutional neural network have emerged as powerful tools for skin cancer detection. Generally, deep learning approaches are computationally expensive and require large storage space. Therefore, deploying such a large complex model on resource‐constrained devices is challenging. An ultra‐light and accurate deep learning model is highly desirable for better inference time and memory in low‐power‐consuming devices. Knowledge distillation is an approach for transferring knowledge from a large network to a small network. This small network is easily compatible with resource‐constrained embedded devices while maintaining accuracy. The main aim of this study is to develop a deep learning‐based lightweight network based on knowledge distillation that identifies the presence of skin cancer. Here, different training strategies are implemented for the modified benchmark (Phase 1) and custom‐made model (Phase 2) and demonstrated various distillation configurations on two datasets: HAM10000 and ISIC2019. In Phase 1, the student model using knowledge distillation achieved accuracies ranging from 88.69% to 93.24% for HAM10000 and from 82.14% to 84.13% on ISIC2019. In Phase 2, the accuracies ranged from 88.63% to 88.89% on HAM10000 and from 81.39% to 83.42% on ISIC2019. These results highlight the effectiveness of knowledge distillation in improving the classification performance across diverse datasets and enabling the student model to approach the performance of the teacher model. In addition, the distilled student model can be easily deployed on resource‐constrained devices for automated skin cancer detection due to its lower computational complexity.

A Topology-Enhanced Multi-Viewed Contrastive Approach for Molecular Graph Representation Learning and Classification.

In recent times, graph representation learning has been becoming a hot research topic which has attracted a lot of attention from researchers. Graph embeddings have diverse applications across fields such as information and social network analysis, bioinformatics and cheminformatics, natural language processing (NLP), and recommendation systems. Among the advanced deep learning (DL) based architectures used in graph representation learning, graph neural networks (GNNs) have emerged as the dominant and highly effective framework. The recent GNN-based methods have demonstrated state-of-the-art performance on complex supervised and unsupervised tasks at both the node and graph levels. In recent years, to enhance multi-view and structured graph representations, contrastive learning-based techniques have been developed, introducing models known as graph contrastive learning (GCL) models. These GCL approaches leverage unsupervised contrastive methods to capture multi-view graph representations by comparing node and graph embeddings, yielding significant improvements in both graph-level representations and task-specific applications, such as molecular embedding and classification. However, as most GCL techniques are primarily designed to focus on the explicit graph structure through GNN-based encoders, they often overlook critical topological insights that could be provided through topological data analysis (TDA). Given the promising research indicating that topological features can greatly benefit various graph learning tasks, we propose a novel topology-enhanced, multi-view graph contrastive learning model called TMGCL. Our TMGCL model is designed to capture and utilize both comprehensive multi-scale topological and global structural information from graphs. This enhanced representation capability positions TMGCL to directly support a range of applications, such as molecular classification, with improved accuracy and robustness. Extensive experiments within two real-world datasets proved the effectiveness and outperformance of our proposed TMGCL in comparing with state-of-the-art GNN/GCL-based baselines.