The rapid growth of generative artificial intelligence (GenAI) has created a significant challenge for the trustworthiness of digital evidence. Artifacts like chat logs, screenshots, document images, and PDF files can now be created or altered with such realism that traditional forensic validation methods, which rely on visual checks, metadata verification, and standard integrity controls, are less effective. This paper introduces EvidenceFusion-MM, a multimodal forensic framework designed to assess evidence integrity across text, document images, and PDF formats. The system combines a transformer-based DistilBERT classifier to detect AI-generated text, a CNN-based Tiny U-Net model for identifying document-image tampering, and an Isolation Forest model for analyzing structural anomalies in PDFs. These specificbranches work together in a single decision pipeline to provide an overall assessment of evidence of integrity risk. Experimental results show strong performance in the textual and PDF areas. The combined text model achieved 96.0% accuracy and a 96.0% macro F1-score, while the PDF branch successfully distinguished between the reference and comparison document sets by flagging 16 out of 300 CompanyDocuments PDFs and 299 out of 300 PDF files as unusual. The document-image branch also proved its ability to analyze localized tampering, achieving a mean IoU of 0.0903 under asimple baseline setting. These results emphasize the importance of multimodal forensic methods in enhancing the reliability of digital evidence in the age of GenAI.

The study introduces a novel approach that uses custom payloads to examine the effectiveness of manual and automated malware detection techniques. The dynamic blended approach can effectively generate attack payloads and identify vulnerabilities in a website, ultimately strengthening the security measures of the websites. This research focuses on dynamic analysis in a controlled environment while testing and analyzing SQL and XSS payloads under varying security conditions. This quantitative analysis involves crafting targeted payloads to bypass Web Application Firewall (WAF) filters and security controls, considering the detection accuracy, response times, false positives and false negatives. The study findings reveal that basic manual payloads are effective against lower security levels, while customized advanced or adaptive payloads could successfully evade detection using obfuscation or encoding techniques. Additionally, in various security constraints, time-based and error-based SQL injections demonstrated significant effectiveness in detection. Automatic scanning using OWASP ZAP identifies common vulnerabilities, but is often ineffective during complex or obfuscated injections. This research underscores the importance of hybrid or blended approaches that combine automated and customized payload techniques for the detection of advanced malware attacks to enhance web application security. This integration of custom payload design with manual and automated scanning approaches can improve detection accuracy and speed up the investigation process to mitigate zero-day vulnerabilities.

Insider threats continue to pose a significant cybersecurity concern due to the inherent trust and authorized access granted to individuals within an organization. Unlike external attacks, insider threats are often more difficult to detect because they involve users operating with legitimate credentials, making their actions appear routine or authorized. This research presents a forensic framework for detecting insider threats by analyzing behavioral patterns in user email activity. Drawing on the CERT Insider Threat Email dataset, a three-phase process which includes time-based activity analysis, feature extraction, and hybrid anomaly detection was implemented with Isolation Forest and Local Outlier Factor (LOF) algorithms, supplemented by Random Forest classification for user-level evaluation. The system detected over 26,000 anomalous events, with a significant proportion occurring outside standard business hours, indicating possible policy violations or misuse of access. An NLP and sentiment analysis layer was integrated to extract contextual insights from anomalous emails, revealing thematic patterns and potential threat-related language. Results demonstrated strong agreement between Isolation Forest and LOF outputs, highlighting the framework’s robustness. While effective in identifying behavioral anomalies without labeled data, limitations included a lack of role-based context and intent determination. Future improvements involve integrating role-informed modeling, and hybrid supervised and unsupervised approaches to enhance accuracy, interpretability, and real-time applicability in operational environments.

Daily software-vulnerability disclosures reported in the National Vulnerability Database (NVD) drive patch management, triage prioritization, and organizational cyber-risk workflows. Yet the short-term dynamics of disclosure activity remain poorly understood and are rarely modeled as a forecasting problem, even though anticipating tomorrow’s disclosure volume can improve operational readiness. This paper develops a cross-domain machine-learning framework for one-step-ahead forecasting of global vulnerability disclosures. The framework integrates technical cybersecurity indicators, cyber-related global news activity, financial-market volatility, macroeconomic conditions, and rich temporal structure through an extensive feature-engineering pipeline. Using a harmonized daily panel covering 2015–2025 [1], [2], we construct two model-ready feature sets that include calendar and seasonal components, lagged variables, rolling-window statistics, realized volatility, entropy measures, jump indicators, and fractal-dimension estimates [3]–[5]. We evaluate simple time-series baselines, regularized regressors, support vector regression, k-nearest neighbors, and several tree-based ensembles under a strictly chronological train–validation–test protocol with rolling-origin cross-validation. The results show that nonlinear ensemble models—especially Gradient Boosting and ExtraTrees—consistently outperform simple baselines and kernel methods, capturing meaningful short-term structure in the NVD disclosure process. Furthermore, a filtered feature subset with reduced dimensionality achieves performance comparable to the full engineered space, indicating that predictive signal arises from a focused combination of temporal, cyber, and cross-domain indicators rather than from sheer feature count. The findings demonstrate that next-day vulnerability-disclosure forecasting is feasible and can support proactive vulnerability management, SOC staffing, and cyber-risk analytics by providing advance warning of elevated disclosure workloads.

We present a probabilistic modeling framework for quantifying decryption-failure probability (DFP) in CRYSTALS-Kyber, the lattice-based key encapsulation mechanism standardized by National Institute of Standards and Technology as module-lattice-based key-encapsulation mechanism. Our method combines exact tail computation of Kyber's centered-binomial noise distribution using FFT-based convolution with principled comparisons to classical concentration inequalities such as Hoeffding and Bernstein. This hybrid analytical-computational approach yields implementation-independent upper bounds on DFP that are exponentially small in the security parameter. Specifically, we compute two-sided tails for aggregated noise variables, translate those into per-ciphertext failure probabilities through a transparent union bound, and determine the minimal reconciliation margins required to ensure DFP <= 2-lambda for lambda is an element of{128,192,256}. Across Kyber-like parameter regimes, Bernstein-type inequalities consistently overestimate risk compared to the exact probabilistic tails, which are several orders of magnitude smaller. The resulting gap highlights the conservatism of inequality-based analyses and clarifies the quantitative safety margins inherent to Kyber's design. These findings contribute to the broader effort of modeling reliability in post-quantum cryptographic primitives using probabilistic and computational methodologies.

The surge of fake accounts on social media platforms and the dissemination of malicious content poses significant cyber threats and escalation of misinformation to online communities. This study focuses on the detection of fake Twitter/X accounts along with identifying malicious URLs distribution from those accounts, utilizing advanced deep learning techniques. The approach employs feature engineering with the Twitter account metadata and user behavioral pattern with a diverse dataset. Deep learning architecture includes a Convolutional Neural Network (CNN) for classifying genuine and fake accounts. Subsequently, an algorithm is developed to find all the URLs from tweets and utilizes a URL-finder service for detecting potential malicious URLs. Various evaluation metrics such as accuracy, precision, recall and F1 score are employed to measure the performance of the deep learning model. Rigorous testing with the proposed model demonstrates robust performance with an accuracy of 95

This research investigates a novel hybridization strategy between Convolutional Neural Networks (CNNs) and gradient-boosted decision trees to enhance image classification accuracy. While conventional approaches focus on optimizing either CNN architectures or machine learning algorithms independently, we propose that intervening in the architecture itself—by strategically replacing the dense classifier portion of the CNN with a tree-based learner—can yield superior results. In our study, we construct a CNN composed of three convolutional blocks, each followed by ReLU activation, max-pooling, and dropout layers. Instead of proceeding through the final dense layers, we extract features immediately after the Flatten layer and input them into an XGBoost classifier. Our experiments reveal that applying XGBoost to these flattened features results in a higher classification accuracy than the fully optimized CNN. Although other datasets were examined during initial testing, this paper focuses exclusively on CIFAR-10 for clarity and reproducibility. The findings suggest that performance gains can be achieved through structural interventions in model architecture, challenging the prevailing emphasis on end-to-end optimization.

Ensuring campus safety is essential for secure and productive learning environments. Although Clery Act Reports provide the most reliable safety data, public rankings often correlate weakly with them, largely due to strong neighborhood influence. A campus with limited internal safety procedures may still receive a high rating if located in a safe area, revealing that current rankings underweight institutional efforts. To address this gap, we apply a data-driven method that classifies campuses based on their safety practices, reported ranks, and surrounding environmental conditions. By integrating Clery Act data into our framework, we highlight institutional safety measures over external factors to provide a more accurate evaluation. Using the extracted dataset, we compute indicators such as the Density of Considered Safety Measures (DCSM) and the Rank-Based Normal DCSM (RND) to measure safety class deviation. These classification results are then used to train an MLP model to predict safety rankings for campuses lacking public indices.

What makes a wine exceptional enough to score a perfect 10 from experts? This study explores a data-driven approach to identify the ideal physicochemical composition for wines that could achieve this highest possible rating. Using a dataset of 11 measurable attributes, including alcohol, sulfates, residual sugar, density, and citric acid, for wines rated up to a maximum quality score of 8 by expert tasters, we sought to predict compositions that might enhance wine quality beyond current observations. Our methodology applies a second-degree polynomial ridge regression model, optimized through an exhaustive evaluation of feature combinations. Furthermore, we propose a specific chemical and physical composition of wine that our model predicts could achieve a quality score of 10 from experts. While further validation with winemakers and industry experts is necessary, this study aims to contribute a practical tool for guiding quality exploration and advancing predictive modeling applications in food and beverage sciences.

The integration of Artificial Intelligence (AI) into project management has significantly transformed traditional methodologies, particularly within the Information Technology (IT) sector. This study investigates the effectiveness of AI-powered tools, specifically Asana, Jira, and Monday.com to improve project management efficiency through automation, resource optimization, and decision-making processes. Using a systematic literature review and quantitative analysis, the research highlights the capabilities and limitations of these tools in the management of complex IT projects. The study findings reveal that AI-powered platforms facilitate real-time tracking, predictive analytics and dynamic resource allocation, which collectively contribute to reduced project timelines and improved outcomes. However, challenges such as data quality, integration complexities, and the need for skilled personnel remain prevalent. The study was summarized with strategic recommendations for organizations to incorporate automation, leverage predictive analytics, and improve data integrity, thus maximizing the benefits of AI in project management. This research underscores the critical role of AI in modernizing project management practices, which ultimately leads to increased productivity and project success rates in the IT domain.

Technological advancements are reshaping our world, bringing with them new cyber threats, including sophisticated adversarial attacks on machine learning models. This research focuses on two types of these attacks: clean-label poisoning and decision-based attacks, which threaten the integrity of models. By utilizing two prominent datasets, MNIST and CIFAR-10, the study performs a comparative analysis to assess how these attacks impact neural network models tailored for image classification. Employing black-box attack techniques, the research simulates realistic conditions that attackers might use. The implementation and evaluation of these models are conducted using TensorFlow, focusing on the effects of attacks on model accuracy and exploring potential defense mechanisms. The objective is to understand the severity of these adversarial attacks and to identify effective defenses to protect machine learning models from such vulnerabilities.

Ensuring robust security and reliability is crucial in the dynamic realm of digital transactions and decentralized systems. This project aims to reshape blockchain transaction security, fostering innovation and trust in decentralized networks. Integrating advanced technologies and sophisticated authentication methods introduces a framework to enhance blockchain transaction security effectively. This framework, which is both practical and efficient, promises unparalleled efficacy in bolstering blockchain networks' integrity, privacy, and trustworthiness. Our research combines blockchain's transparency and immutability with DID systems' cryptographic capabilities and ID.me's user-centric authentication features, setting a new standard for blockchain security. Our study demonstrates the practicality and efficiency of this framework through experimentation and analysis, with profound implications for a wide range of industries. Positioned to shape the future of secure blockchain transactions, our study heralds a new era of innovation and trust in decentralized networks, with far-reaching benefits for finance, healthcare, supply chain management, and beyond.

Drones are integral to various applications, out of which traffic surveillance is an important application. However, their operational efficiency is limited by battery life, which restricts their capacity for extended critical missions. Additionally, in remote or high-interference areas, the bandwidth for drone communication is often limited, leading to a decrease in the quality of images transmitted to the base station. This paper aims to address such challenges by having drones transmit video data in real-time at lower resolutions for traffic monitoring. This approach conserves energy and optimizes transmission. However, it adversely affects object detection accuracy at the base station due to compromised data quality. To address this issue, we incorporate Generative Adversarial Networks (GANs) to improve LR images, restoring their quality for precise object detection. Results indicate that the accuracy of traffic analytics achieved with GAN-enhanced images is comparable to that obtained with high-resolution data transmission. Consequently, our approach allows a fundamental trade-off among drone energy consumption, transmission time, flight time, and object detection accuracy, enabling robust detection performance while conserving energy and enhancing operational capabilities.

Our research introduces a novel method for determining the originating software of digital images, which significantly advances digital forensic analysis capabilities. This method involves transforming images into their hexadecimal code representations, thereby stripping away metadata and making the files unrecognizable by conventional identification techniques. Through a meticulous analysis of these hex codes, broken down into 2-character substrings, we construct detailed feature vectors representing the frequency of each substring. Utilizing a diverse array of machine learning models, including RandomForestClassifier, LogisticRegression, and others, our approach successfully identifies the software used to create the images, such as PowerPoint, GIMP, Picasa, and the online tool Batchtools.pro, with an impressive accuracy rate between 97

Adversarial training has emerged as a promising defense mechanism involving the addition of adversarial examples into training data to enhance model robustness. Despite its potential, research on adversarial machine learning attack and resistance methods in the digital domain remains limited, requiring further investigation into its efficacy, specifically in defending against misclassification in smart vehicles. This study investigates the application of adversarial training to combat misclassification errors in smart vehicles, highlighting its necessity due to these systems' vulnerability to adversarial machine learning attacks, especially through their connection to external networks like the Controller Area Network. This paper also analyzes the significant challenges, defense mechanisms, and vulnerabilities specific to smart vehicle systems, offering insights to guide future research and improve the reliability of autonomous driving technologies.

The integration of the Internet of Things (IoT) into modern life has become omnipresent in the personal and business domains. IoT devices, which collect and communicate data for various services, raise privacy concerns as they infect all aspects of contemporary life. This survey examines the privacy and security challenges posed by IoT devices, particularly in smart homes, industrial settings, and healthcare settings. Rapid development of IoT hardware and software often prioritizes functionality over security, leading to vulnerabilities that can compromise personal and organizational privacy. In severe cases, security breaches in IoT devices can result in physical harm. This study explores the unique security needs of Smart Homes, Industrial IoT (IIoT), and Internet of Medical Things (IoMT), proposing that solutions developed for one domain may be applicable to others due to the heterogeneous nature of IoT.

As businesses become more digitalized, the software supply chain (SSC) has grown increasingly vital, yet it's also becoming more vulnerable to cyber threats. This is evident in the recent surge in attacks targeting the SSC. Highlighting the indispensable need for robust SSC security measures, this research emphasizes the importance of securing every facet of the software supply chain to mitigate unauthorized access, data breaches, and the insertion of malicious code. We advocate for the integration of Security Orchestration, Automation, and Response (SOAR) solutions, particularly through low-code or no-code platforms like Robotic Process Automation (RPA) to enhance Security Information and Event Management (SIEM) systems. RPA's automation capabilities enhance cybersecurity efficiency by streamlining routine security tasks, thereby enabling Security Operations Center (SOC) analysts to focus on strategic security initiatives. Inspired by the MOVEit Transfer cyberattack incident of 2023, this research underscores the financial and operational impacts of SSC vulnerabilities and presents the application of RPA to automate two specific tasks highlighted from the analysis of the MOVEit Transfer attack. We put these tasks through rigorous testing and demonstration to showcase their integration potential for an organization. Our work not only exhibits the cost-effectiveness and scalability of RPA in bolstering security measures but also emphasizes its role in expediting threat detection and response. This research concludes with a proposed automation approach for SSC security enhancement, detailing the application of RPA in automating specific security tasks, thereby advocating for a more resilient, efficient, and adaptable security infrastructure in the face of evolving cyber threats.

Earthquakes pose a significant threat to urban areas, necessitating accurate forecasting models to mitigate their impact. This study focuses on earthquake forecasting in Los Angeles, a region with high seismic activity and limited research. We established a feature matrix for forecasting earthquakes within a 30-day period by analyzing the most predictive patterns from recent studies. Our model developed a subset of features capable of forecasting the highest magnitude of an earthquake. Using advanced machine learning algorithms and neural networks, our model achieved an accuracy of 69.14% in forecasting the maximum magnitude earthquake as one of the 6 categories. We aim to provide a useful guideline for future researchers.

When the ground communication base stations in the target area are severely destroyed, the deployment of Unmanned Aerial Vehicle (UAV) ad hoc networks can provide people with temporary communication services. Therefore, it is necessary to design a multi-UAVs cooperative control strategy to achieve better communication coverage and lower energy consumption. In this paper, we propose a multi-UAVs coverage model based on Adaptive Virtual Force-directed Particle Swarm Optimization (AVF-PSO) strategy. In particular, we first introduce the gravity model into the traditional Particle Swarm Optimization (PSO) algorithm so as to increase the probability of full coverage. Then, the energy consumption is included in the calculation of the fitness function so that maximum coverage and energy consumption can be balanced. Finally, in order to reduce the communication interference between UAVs, we design an adaptive lift control strategy based on the repulsion model to reduce the repeated coverage of multi-UAVs. Experimental results show that the proposed coverage strategy based on gravity model outperforms the existing state-of-the-art approaches. For example, in the target area of any size, the coverage rate and the repeated coverage rate of the proposed multi-UAVs scheduling are improved by 6.9–29.1%, and 2.0–56.1%, respectively. Moreover, the proposed scheduling algorithm is high adaptable to diverse execution environments.© 2022 Published by Elsevier Ltd.