Socially Aware Synthetic Data Generation for Suicidal Ideation Detection Using Large Language Models

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 14350 - 14363

Published: Jan. 1, 2024

Suicidal ideation detection is a vital research area that holds great potential for improving mental health support systems. However, the sensitivity surrounding suicide-related data poses challenges in accessing large-scale, annotated datasets necessary training effective machine learning models. To address this limitation, we introduce an innovative strategy leverages capabilities of generative AI models, such as ChatGPT, Flan-T5, and Llama, to create synthetic suicidal detection. Our generation approach grounded social factors extracted from psychology literature aims ensure coverage essential information related ideation. In our study, benchmarked against state-of-the-art NLP classification specifically, those centered around BERT family structures. When trained on real-world dataset, UMD, these conventional models tend yield F1-scores ranging 0.75 0.87. data-driven method, informed by factors, offers consistent 0.82 both suggesting richness topics can bridge performance gap across different model complexities. Most impressively, when combined mere 30% UMD dataset with data, witnessed substantial increase performance, achieving F1-score 0.88 test set. Such results underscore cost-effectiveness confronting major field, scarcity quest diversity representation.

Language: Английский

---

A performance overview of machine learning-based defense strategies for advanced persistent threats in industrial control systems

Computers & Security, Journal Year: 2023, Volume and Issue: 134, P. 103445 - 103445

Published: Aug. 24, 2023

Language: Английский

---

LogRF: An Approach to Human Pose Estimation Using Skeleton Landmarks for Physiotherapy Fitness Exercise Correction

IEEE Access, Journal Year: 2023, Volume and Issue: 11, P. 107930 - 107939

Published: Jan. 1, 2023

Human pose and gesture estimation are crucial in correcting physiotherapy fitness exercises. In recent years, advancements computer vision machine learning approaches have led to the development of sophisticated models that accurately track analyze human movements real time. This technology enables physiotherapists trainers gain valuable insights into their client's exercise forms techniques, facilitating more effective corrections personalized training regimens. research aims propose an efficient artificial intelligence method for during We utilized a multi-class dataset based on skeleton movement points conduct our experimental research. The comprises 133 features derived from various exercises, resulting high feature dimensionality affects performance with deep methods. introduced novel Logistic regression Recursive Feature elimination (LogRF) selection. Extensive experiments demonstrate using top twenty selected features, random forest outperformed state-of-the-art studies high-performance score 0.998. each applied is validated through k-fold approach further enhanced hyperparameter tuning. Our proposed study assists specialists identifying addressing potential biomechanical issues, improper postures, incorrect patterns, which essential injury prevention optimizing outcomes. Furthermore, this enhances capabilities remote monitoring guidance capabilities, allowing support patient's progress prescribed exercises continually.

Language: Английский

---

A novel ensemble method for enhancing Internet of Things device security against botnet attacks

Decision Analytics Journal, Journal Year: 2023, Volume and Issue: 8, P. 100307 - 100307

Published: Aug. 23, 2023

The growing number of connected Internet Things (IoT) devices has led to the daily growth network botnet attacks. networks compromised controlled by a single entity can be used for malicious purposes such as denial service distributed IoT attacks and theft personal information. weak security measures many make them easy targets compromise inclusion in botnets. In this research, we propose system detecting We develop an ensemble learning detect botnets traffic with high-performance scores. will analyze identify any suspicious behavior that may indicate presence botnet. For purpose, use benchmark CTU-13 dataset build applied machine deep techniques comparison. novel technique, K-neighbors, Decision tree, Random forest (KDR), achieve high performance attack detection. Study results show proposed KDR gives 99.7% accuracy 12.99 s. Hyperparameter optimization k-fold cross-validation are employed substantiate performance. Our research study contributes body knowledge on detection provides practical solution securing against

Language: Английский

---

Preventing Crimes Through Gunshots Recognition Using Novel Feature Engineering and Meta-Learning Approach

IEEE Access, Journal Year: 2023, Volume and Issue: 11, P. 103115 - 103131

Published: Jan. 1, 2023

Gunshot sounds are common in crimes, particularly those involving threats, harassment, or killing. The gunshot crimes can create fear and panic among victims, often leading to psychological trauma. associated with a significant mortality rate, especially cases of gun violence. sound gunshots serve as evidence criminal investigations, allowing law enforcement officials determine the number shots fired, caliber used, distance from which were fired. Efficient detection is necessary address issue violence society. This study aims detect using an efficient approach prevent crimes. frequency-time domain spectrum analysis performed understand patterns signals related each target class. A novel Discrete Wavelet Transform Random Forest Probabilistic (DWT-RFP) feature engineering proposed, takes Mel-frequency cepstral coefficients (MFCC) extracted data input for extraction. meta-learning-based Meta-RF-KN (MRK) proposed based on newly created ensemble features DWT-RFP approach. For experiments, dataset containing 851 audio clips collected public videos YouTube eight kinds models, used. Advanced machine learning deep techniques applied comparison evaluate performance Extensive experiments show that MRK achieves 99% k-fold accuracy detecting outperforms state-of-the-art approaches. potentially be used accurate help

Language: Английский

---

Novel Sentiment Majority Voting Classifier and Transfer Learning-Based Feature Engineering for Sentiment Analysis of Deepfake Tweets

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 67117 - 67129

Published: Jan. 1, 2024

Deepfake text known as synthetic text, involves using artificial intelligence (AI)-generated to create fabricated information or imitate actual individuals. Twitter tweets related deepfake can be used for many malicious intents, including impersonation, creating fake news, and spreading misinformation. The main goal of this investigation is detect people's sentiments technology with an advanced technique. A novel sentiment majority voting classifier (SMVC) proposed the labeling collected tweets. SMVC selects final from three lexicon-based models TextBlob, valence-aware dictionary reasoner (VADER), AFINN a mechanism. For classification, we propose transfer feature where embedding features are fed long short-term memory (LSTM), decision tree (DT) outputs combined into single set. Extensive experiments show that learning-based engineering results in highest performance. logistic regression outperforms accuracy 98.9% minimum computational complexity. classification performance each applied model validated k-fold cross-validations. Moreover, assessment existing state-of-the-art also carried out robustness

Language: Английский

---

Novel Transformer Based Contextualized Embedding and Probabilistic Features for Depression Detection From Social Media

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 54087 - 54100

Published: Jan. 1, 2024

Depression constitutes a significant mental health condition, impacting an individual's emotional state, thought processes, and ability to carry out everyday tasks. is defined by ongoing feelings of sadness, diminished interest in previously enjoyed activities, alterations hunger, sleep disturbances, decreased vitality, challenges with focus. The impact depression extends beyond the individual, affecting society at large through productivity higher healthcare costs. In realm social media, users often express their thoughts emotions posts, which can provide insightful data for identifying patterns depression. This research aims detect early analyzing media user content machine learning techniques. We have built advanced models using benchmark database containing 20,000 tagged tweets from profiles identified as depressed or non-depressed. are introducing innovative BERT-RF feature engineering method that extracts Contextualized Embeddings Probabilistic Features textual input. Bidirectional Encoder Representations Transformers (BERT) model, based on Transformer architecture, used extract Embedding features. These features then fed into random forest model generate class probabilistic prominent aid enhancing identification media. order classify derived selection step, we five popular classifiers: Random Forest (RF), Multilayer Perceptron (MLP), K-Neighbors Classifier (KNC), Logistic Regression (LR), Long Short-Term Memory (LSTM). Evaluation experiments show our approach, engineering, enables outperform state-of-the-art methods high accuracy score 99%. validated results k-fold cross-validation statistical T-tests. achieved 99% during validation proposed approach. contributes significantly computational linguistics analytics providing robust approach detection content.

Language: Английский

---

CVG-Net: novel transfer learning based deep features for diagnosis of brain tumors using MRI scans

PeerJ Computer Science, Journal Year: 2024, Volume and Issue: 10, P. e2008 - e2008

Published: May 17, 2024

Brain tumors present a significant medical challenge, demanding accurate and timely diagnosis for effective treatment planning. These disrupt normal brain functions in various ways, giving rise to broad spectrum of physical, cognitive, emotional challenges. The daily increase mortality rates attributed underscores the urgency this issue. In recent years, advanced imaging techniques, particularly magnetic resonance (MRI), have emerged as indispensable tools diagnosing tumors. MRI scans provide high-resolution, non-invasive visualization structures, facilitating precise detection abnormalities such This study aims propose an neural network approach Our experiments utilized multi-class image dataset comprising 21,672 images related glioma tumors, meningioma pituitary We introduced novel network-based feature engineering approach, combining 2D convolutional (2DCNN) VGG16. resulting 2DCNN-VGG16 (CVG-Net) extracted spatial features from using 2DCNN VGG16 without human intervention. newly created hybrid set is then input into machine learning models diagnose balanced data Synthetic Minority Over-sampling Technique (SMOTE) approach. Extensive research demonstrate that utilizing proposed CVG-Net, k-neighbors classifier outperformed state-of-the-art studies with k-fold accuracy performance score 0.96. also applied hyperparameter tuning enhance tumor diagnosis. has potential revolutionize early diagnosis, providing professionals cost-effective diagnostic mechanism.

Language: Английский

---

An efficient genetic disorder detection framework using adaptive segmentation and classification mechanism from chromosome images

Expert Systems with Applications, Journal Year: 2025, Volume and Issue: unknown, P. 127303 - 127303

Published: March 1, 2025

Language: Английский

---

Federated Multi-Label Learning (FMLL): Innovative Method for Classification Tasks in Animal Science

Animals, Journal Year: 2024, Volume and Issue: 14(14), P. 2021 - 2021

Published: July 9, 2024

Federated learning is a collaborative machine paradigm where multiple parties jointly train predictive model while keeping their data. On the other hand, multi-label deals with classification tasks instances may simultaneously belong to classes. This study introduces concept of Multi-Label Learning (FMLL), combining these two important approaches. The proposed approach leverages federated principles address tasks. Specifically, it adopts Binary Relevance (BR) strategy handle nature data and employs Reduced-Error Pruning Tree (REPTree) as base classifier. effectiveness FMLL method was demonstrated by experiments carried out on three diverse datasets within context animal science: Amphibians, Anuran-Calls-(MFCCs), HackerEarth-Adopt-A-Buddy. accuracy rates achieved across were 73.24%, 94.50%, 86.12%, respectively. Compared state-of-the-art methods, exhibited remarkable improvements (above 10%) in average accuracy, precision, recall, F-score metrics.

Language: Английский

---