Journal of Computers, Mechanical and Management

Detecting Depression Using Twitter Data by Incorporating Hybrid Feature Representation

2025-12-09T20:48:59+03:00

Depression is a critical global mental health challenge that often remains undiagnosed due to the limitations and subjectivity of conventional screening techniques. The growing use of social media platforms offers new avenues for understanding human emotions, as individuals increasingly share their thoughts, moods, and experiences online. Leveraging this vast digital footprint, the present study introduces a machine learning (ML)-driven approach for the automated detection of depression using Twitter data. A comprehensive dataset comprising 205,271 posts was collected and carefully preprocessed through multiple natural language processing (NLP) techniques, including tokenization, stop-word elimination, lemmatization, and sentiment polarity assessment, to extract meaningful textual features. Six distinct ML models were trained and evaluated: Support Vector Classifier (SVM), Logistic Regression, Decision Tree, AdaBoost, Na"{i}ve Bayes, and K-Nearest Neighbors (KNN). Various performance metrics, including accuracy, precision, recall, and F1-score, were employed to assess the efficiency of each developed model. Among the tested models, Logistic Regression achieved the highest accuracy (92%), followed by SVM with 90%, while KNN performed comparatively lower with 70%. The results indicate that linear and ensemble-based classifiers are more effective than distance-based models in managing high-dimensional text data. Overall, this study offers a robust comparative evaluation of ML algorithms for depression detection and underscores the transformative potential of NLP and social media analytics in scalable, data-driven mental health monitoring systems.

Predictive Analytics Model for AI-Enhanced Decision Support in Corporate Management

2025-12-03T10:42:59+03:00

AI and predictive analytics have revolutionized corporate management by replacing guesswork with facts. A comprehensive literature review reveals that AI-enhanced decision support systems are increasingly incorporating machine learning predictive analytics models. This paper summarizes research from academic studies and case studies performed by businesses to demonstrate how predictive analytics becomes an integrated part of planning corporate strategy, allocating resources among departments, and ensuring administrative efficiency. This study focuses on classifiers, which are fundamental machine learning techniques for predicting and simplifying complex decision-making processes. Such techniques include neural networks, regression analysis, decision trees, and others. The authors explain the various aspects to consider when implementing AI-driven solutions successfully, including data quality, model interpretability, and ethics. The findings show that organizations adopting predictive analytics report measurable improvements, including up to 15% reduction in employee turnover, 20 -30% improvement in risk mitigation, 25% sales growth, and 40% reduction in operational inefficiencies when integrated with ERP systems. The study also examines how predictive analytics is affecting various disciplines, such as risk management, market trend forecasting, and employee performance appraisal, by analyzing specific real-life examples. Findings suggest that the implementation of real-time analytics in ERP systems has the potential to enhance strategic decision-making significantly. The review also reveals gaps in the literature and contributes to future research by highlighting the need to scale solutions to problems and applications across industries.

Leveraging Cloud Computing for Real-Time Big Data Analytics in Healthcare Systems

2025-12-03T07:43:35+03:00

Electronic health records (EHRs), medical imaging, and wearable devices generate vast amounts of data in healthcare systems, necessitating scalable, real-time analytics. The proposed study recommends a cloud-based system that combines streaming ingestion and Apache Spark processing with machine learning-based predictive models and Gomoku-inspired stream-cipher encryption, along with blockchain validation. The proposed framework is designed to support continuous data ingestion, low-latency processing, and secure analytics in distributed healthcare environments. Experimental evaluation demonstrates that the system achieves a 96.5% F1-score, 120 ms processing time, 450 MB/s throughput, and 68% resource usage, outperforming Hadoop (90.2% F1-score, 350 ms) and Spark (94.1%, 200 ms) in processing latency, throughput, and resource efficiency. Findings further indicate that operational costs would be reduced by 30%, early intervention rates would improve by 85%, and patient engagement would increase by 40%, enabling the delivery of proactive and personalized healthcare. Additionally, integrating game-theoretic key generation with cloud-based analytics enhances data confidentiality and integrity for real-time healthcare applications. The proposed approach demonstrates the feasibility of combining scalable cloud computing, real-time big data analytics, and advanced security mechanisms to address the growing demands of modern healthcare systems.

A Multidimensional Evaluation Framework for Parallel Frequent Pattern Mining Algorithms on Big Data Platforms

2025-11-26T07:45:57+03:00

Despite ongoing theoretical refinements in parallel frequent pattern mining algorithms, practical implementations still experience issues such as inefficient resource scheduling, low node interaction efficiency, and limited system robustness. Addressing the lack of comprehensive and systematic testing methodologies for existing algorithms, this paper proposes a data mining algorithm testing framework tailored for big data platforms. The methodology centers on three key dimensions: resource adaptability, communication efficiency, and system robustness, establishing a quantifiable and reproducible experimental evaluation framework. To validate its effectiveness, KVBFP is used as the experimental subject, and three sets of experiments are designed and implemented within a Hadoop cluster environment: algorithm resource adaptability testing, communication frequency testing, and algorithm stability testing. Experimental results show that the first set of experiments accurately measures algorithm resource consumption across different clusters. The second set of experiments shows that KVBFP reduces communication frequency by 51.5% compared to the PFP algorithm. The third set of experiments demonstrates that the algorithm’s recovery time remains within 30 seconds under fault conditions. Through comprehensive evaluation of the algorithm via these three experiments, this paper provides a quantitative reference for applying data mining algorithms in real-world scenarios.

A Comprehensive Review of Artificial Intelligence for Image- and Signal-Based Nondestructive Testing in Aerospace Structures

2025-12-03T07:55:54+03:00

Ensuring the structural integrity of aerospace components requires inspection techniques that can detect diverse surface and subsurface flaws in increasingly complex materials and geometries. Although conventional nondestructive testing (NDT) remains essential, its dependence on manual interpretation and limited automation has created demand for more objective, scalable solutions. This review presents a structured synthesis of artificial intelligence advancements in nondestructive testing, organized around the two dominant data paradigms: image-based and signal-based inspection. Image modalities, such as radiography, infrared thermography, and visual inspection, generate spatial information well-suited to convolutional networks, segmentation models, and vision transformers. Signal modalities, including ultrasonics, acoustic emission, eddy currents, and vibration analysis, produce temporal or spectral data that can be effectively modeled by recurrent neural networks (RNNs), hybrid CNN-LSTM architectures, and emerging transformers. The review compares these modalities, evaluates their diagnostic performance, and highlights challenges related to dataset scarcity, inconsistent annotation standards, domain shift, interpretability, and certification. Particular attention is given to multimodal fusion strategies that integrate spatial and temporal cues through attention-enabled hybrid models to improve robustness and decision reliability. Practical aerospace scenarios such as composite panel inspection, ultrasonic C-scan analysis, radiographic porosity detection, and structural health monitoring are examined to illustrate operational readiness. Despite significant progress, most models rely on controlled datasets, lack standardized evaluation protocols, and provide limited insight into uncertainty or failure modes. Advancements in open benchmarks, explainable and physics-informed architectures, and digital-twin-enabled deployment are essential for achieving trustworthy, certifiable AI-based NDT. Overall, the review provides a concise roadmap for developing intelligent and interpretable NDT systems for next-generation aerospace applications.

A Systematic Review of Scalable Blockchain-Based Digital Signature Frameworks for Healthcare Data Security

2025-10-05T07:30:46+03:00

This is the first PRISMA-guided systematic review of scalable blockchain digital signatures for healthcare, synthesizing evidence from 85 peer-reviewed studies published between 2015 and 2024. The review examines five thematic areas: digital signatures, consensus mechanisms, smart contracts, hybrid blockchain architectures, and regulatory compliance. Particular emphasis is placed on scalability challenges and the role of alternative consensus protocols, such as Proof-of-Stake and Delegated Proof-of-Authority (DPoA), in addressing the energy and latency limitations of Proof-of-Work (PoW). Findings highlight the value of smart contracts in automating consent and authentication processes, while hybrid blockchain models are shown to balance security with scalability. The synthesis also identifies persistent challenges, including interoperability with legacy systems, energy consumption, and compliance with GDPR and HIPAA regulations. Importantly, emerging approaches such as Layer-2 scaling, AI-enhanced validation, and post-quantum cryptography are highlighted as promising directions. By integrating technical and regulatory perspectives, this review contributes a critical roadmap for researchers, healthcare providers, and system architects seeking secure, efficient, and regulation-compliant blockchain frameworks

Editorial Comments

2025-12-28T07:27:47+03:00