Prompt-based contrastive learning to combat the COVID-19 infodemic

Machine Learning, Journal Year: 2025, Volume and Issue: 114(1)

Published: Jan. 1, 2025

Language: Английский

Explainable TabNet ensemble model for identification of obfuscated URLs with features selection to ensure secure web browsing

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: March 19, 2025

Obfuscated and malicious URLs may lead to harmful content or actions the system, such as downloading malware, phishing, scams, adware. In domain of cybersecurity, identification obfuscated Uniform Resource Locator (URL) is a concerning facet. This study proposes Robust unified TabNet ensemble model for Malicious with feature extraction based on computation features' importance classification. A fine-tuned attention-based deep neural network used extract features URL. The customized data most important generated, Machine Learning (ML) developed classification URLs. evaluation parameters accuracy, Precision, Recall, F1-score are measured look at performance model. Accuracy 97.8%, precision 0.978, recall 0.976, 0.978 reflect outperforming results proposed while classifying five URL classes. further validated through statistical analysis by measuring Kappa value, which comes up 0.968 With 10-fold cross-validation model, we attained mean accuracy 97.27% confidence interval 0.004. Local Interpretable Model-agnostic Explanations (LIME) explainable AI validate perceive contributing towards compared state-of-the-art ML classifiers previous studies, whole validation process favors model's efficacy.

Language: Английский

Integrating Statistical Methods and Machine Learning Techniques to Analyze and Classify COVID-19 Symptom Severity

Big Data and Cognitive Computing, Journal Year: 2024, Volume and Issue: 8(12), P. 192 - 192

Published: Dec. 16, 2024

Background/Objectives: The COVID-19 pandemic, caused by Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2), led to significant global health challenges, including the urgent need for accurate symptom severity prediction aimed at optimizing treatment. While machine learning (ML) and deep (DL) models have shown promise in predicting using imaging clinical data, there is limited research utilizing comprehensive tabular datasets. This study aims address this gap leveraging a detailed dataset develop robust categorizing severity, thereby enhancing decision making. Methods: A unique was created questionnaire responses from 5654 individuals, demographic information, comorbidities, travel history, medical data. Both unsupervised supervised ML techniques were employed, k-means clustering categorize into mild, moderate, severe clusters. In addition, classification models, namely, Support Vector Machine (SVM), Adaptive Boosting (AdaBoost), eXtreme Gradient (XGBoost), random forest, neural network (DNN) used predict levels. Feature importance analyzed forest model its robustness with high-dimensional data ability capture complex non-linear relationships, statistical significance evaluated through ANOVA Chi-square tests. Results: Our showed that fatigue, joint pain, headache most important features severity. SVM, AdaBoost, achieved an accuracy of 94%, while XGBoost 96%. DNN performance handling patterns 98% accuracy. terms precision recall metrics, both demonstrated performance, particularly moderate class. recorded 97% recall, 99% recall. approach improved reducing noise dimensionality. Statistical tests confirmed additional like Body Mass Index (BMI), age, dominant variant type. Conclusions: Integrating advanced offers promising classification. method provides reliable tool healthcare professionals optimize patient care resource management, managing potential future pandemics. Future work should focus on incorporating further enhance applicability.

Language: Английский