This paper presents a telemetry dataset capturing resource utilization and power consumption metrics across the ENACT edge-cloud continuum. The dataset contains empirical telemetry collected in real-time for both infrastructure nodes and application workloads. More specifically, a distributed weather forecasting scenario has been emulated, comprising five pods: two different weather data sources, two forecasting services (one per node/computing layer) and one long-term storage pool. A cloud-based machine and an edge device belonging to the same Kubernetes cluster have been considered for the deployment of the application pods, corresponding to heterogeneous computing tiers. Data acquisition was performed using ENACT's Telemetry Data Collector and Monitoring Engine which measures telemetry and energy metrics at node and pod levels in real-time. The resulting dataset provides time-series records including CPU, memory and disk utilization, network throughput, and energy consumption for the cloud node, the edge node and the five application pods. Telemetry data was collected during two distinct phases: for a period with application workloads running normally and for a baseline period when applications were removed from the cluster. This allows for assessing the impact of the applications activity in terms of resource usage and energy consumption. This dataset offers valuable insights for the research community in distributed systems, the edge-cloud continuum and cognitive computing, wherein datasets on real-world data, especially reflecting both infrastructure-level and application-level telemetry, are currently very limited. It is particularly useful for developers and research scientists that require such data for tasks such as training and fine-tuning time-series forecasting models, benchmarking anomaly detection models and validating scheduling algorithms and energy-aware strategies, to name a few.

Habitat selection studies in large mammals typically rely on photo-interpreted forest maps to link the telemetry locations of individuals to environmental data. Such forest maps mainly provide information on forest composition, age, and disturbance history but say little about the structure of stands. In contrast, airborne LiDAR (Light Detection and Ranging) can provide 3D metrics of vegetation structure, but its application to the study of wildlife–habitat relationships remains limited. We aim to determine if combining these products could improve our capacity to understand the habitat selection patterns of large mammal species representative of the eastern Canadian boreal forest: caribou ( Rangifer tarandus caribou ), moose ( Alces alces americana ) and eastern coyote ( Canis latrans ). We built resource selection functions with mixed logistic regressions to characterize habitat selection patterns, using telemetry data and the different sources of information on forest composition and structure. We evaluated model performance with a k -fold cross-validation. Our results suggest that integrating LiDAR data with forest maps substantially improves the ability to characterize habitat selection patterns across species, though benefits varied with periods and study areas. For example, vegetation structure, mostly detailed by LiDAR data, was the main determinant for caribou and eastern coyotes in the snow-covered period, whereas forest composition, described in the forest maps, was most important to characterize habitat selection patterns for moose in both periods. These differences may be partly attributed to contrasting compositions between northernmost and southernmost forests in our study area (i.e. province of Quebec), species ecology, as well as potential temporal discrepancies between data sources. When used appropriately, combining LiDAR with traditional forest maps provides richer ecological insight and a more comprehensive characterization of habitat selection patterns of large boreal mammals. Based on an average population-level description of habitat selection patterns, this integrated approach can guide practitioners to preserve sparse understory to support caribou, promote complex shrub structure for moose, and limit dense cover that may favor eastern coyotes. • Forest maps inform on composition and disturbance while LiDAR captures structure. • We combined LiDAR and forest maps to model habitat selection for three species. • We compared model parsimony and evaluated performance with k -fold cross-validation. • Combining data sources outperformed simpler models across species and seasons. • Considering stand structure variables could improve forest management for wildlife.

In spatial ecology, the concept of resource selection expresses the idea that for many animals, the distribution of an individual's location is not uniform over the region available to them; instead, they spend time preferentially in some locations compared to others, in a way that can often be related to spatial covariates. The related concept of step selection describes the variation in an individual's tendency to move to particular locations in the short term, taking into account both spatial covariates and the constraints of their process of movement. Consistent modelling of resource selection and step selection is necessary to understand animals' distribution in space and to interpret movement, telemetry, and spatial survey data in a meaningful way. In this paper, I take advantage of recent developments in stochastic processes and statistical algorithms to develop a range of new stochastic models in which both the dynamics and the long-term behaviour are tractable and described parametrically, and which are flexible enough to represent a wide range of patterns of movement and space use encountered in reality. I extend the mathematical analogy between movement modelling and Markov chain Monte Carlo algorithms, first proposed by Michelot, Blackwell & Matthiopoulos (2019; Ecology 100, e02452), to a wide range of continuous-time stochastic processes, including both diffusion processes and velocity-jump models, that in different ways are motivated by the simple discrete-time step-and-turn models widely used in practice. Particular cases include a diffusion process where the dynamics are defined in terms of speed and direction of movement, and a velocity-jump process in d dimensions, generalizing the 'bouncy particle sampler' used in Bayesian inference, in which the distribution of velocity after a so-called 'bounce' event has support over a region which itself has dimension d. I also show how this mathematical approach can be extended to models incorporating distinct behavioural states and to higher dimensional models representing the joint movement of interacting individuals.

Wild animals can adapt to the increasing presence of humans by either becoming accustomed to it or by avoiding humans by spatiotemporal separation. The return of the wolf to the German lowlands raised the opportunity to study wolf behaviour in one of the most densely populated and fragmented countries in Europe, in an area where topography offers no retreat from human disturbance. We analysed telemetry data of 18 wolves from five federal states in Germany to study how wolves adjust their movements and activity in relation to anthropogenic structures. We found no evidence that wolves in Germany have become accustomed to humans or human infrastructure. Our results show that wolves adapted their spatiotemporal behaviour in a way that minimized encounters with humans: 1) wolves were mostly inactive during the day, when humans are most active; 2) they showed a high preference for habitat that provides cover, especially during daylight; and 3) they strongly avoided human infrastructures, especially during daylight. This study shows that land‐sharing between wolves and humans does not appear to have resulted in a loss of strong spatial avoidance of humans, particularly during daytime when human activity is highest. It reaffirms that coexistence in landscapes heavily impacted by humans is possible.

Domestic animals represent the largest mammalian biomass on Earth, creating an urgent need to understand their effects on native species. Environmental change may further alter these interactions, particularly for large herbivores that are highly sensitive to climate-driven resource availability. Interactions among assemblages of large herbivores include interference and exploitation competition, but the grazing optimization hypothesis paradigm suggests that interactions are often facilitative because defoliation from grazing stimulates compensatory regrowth of high-quality forage that benefits other species. We used 35 years of contemporaneous telemetry data on elk and domestic cattle to quantify spatial interactions within the context of climate-change-induced drought conditions. We fit step-selection functions to assess the immediate response of elk to spatiotemporal cattle distribution and tested whether drought and phenological advancement of the growing season-mediated interactions. We also modelled elk selection for pastures that were ungrazed, currently grazed or previously grazed by cattle to assess competition and facilitation. Our results refute the grazing optimization hypothesis and suggest elk are not attracted to areas previously grazed by cattle. Instead, elk avoided cattle throughout the entire growing season and at multiple scales. Elk were 2.9 times more likely to make movements to locations without cattle versus locations with 1 cow-calf pair/hectare, 23% less likely to use a pasture with cattle present, and 27% less likely to use one grazed by cattle earlier in the season, compared to an ungrazed pasture. After controlling for current season grazing status, elk were 30% less likely to use a pasture if it was grazed the previous year. Resource availability, influenced by growing season phenology and drought, mediated the magnitude of avoidance of cattle by elk. In the periods of most severe drought, elk reduced their avoidance of cattle ostensibly to seek out the remaining palatable forage regardless of cattle presence. Increased overlap between domestic and wild ungulates should be anticipated as climate change continues to increase the intensity and frequency of drought, which may result in decreased weight gain or fitness consequences to both species, increased potential for disease transmission, and more encounters between predators and domestic livestock.

Explainable artificial intelligence (XAI) is increasingly required for anomaly detection in high-dimensional sensor systems operating in safety-critical and resource-constrained environments. While existing post-hoc explanation methods provide useful insights, they often suffer from high computational cost, unstable attributions, and limited applicability in unlabeled or unsupervised settings. This paper proposes KFASL, a variance-stable and computationally efficient XAI framework that approximates Shapley-based feature attributions using a variance-optimized weighting strategy. The framework integrates local and global explanations with causality-aware regularization to improve attribution stability and interpretability under limited labeling conditions. The proposed approach reduces the computational complexity of Shapley approximation from exponential to polynomial time, enabling scalable deployment for high-dimensional telemetry data. KFASL is evaluated using a combination of real and synthetic datasets, with spacecraft telemetry used as a representative safety-critical case study. Experimental results demonstrate improved attribution stability, explanation fidelity, and runtime efficiency compared to existing XAI techniques, including SHAP, Kernel SHAP, LIME, and Anchors. These results indicate that KFASL provides a general and practical solution for explainable anomaly detection in complex sensor-driven systems.

Global Navigation Satellite Systems (GNSS) serve as a fundamental tool in modern movement ecology; however, the transnational use of electronic warfare (EW) systems poses a critical threat to telemetry research. Targeted jamming and spoofing of navigation signals result in massive spatiotemporal track distortions, rendering raw data unsuitable for population and spatial analyses. This article analyzes existing data-cleaning tools (in R and Python) and presents three author-developed cascading algorithms (in Python) for automated anomaly identification and true trajectory reconstruction. The proposed methodology integrates deterministic kinematic heuristics (iterative filters for speed and turning angle, and an adaptive spatial deviation) with unsupervised (Isolation Forest) and supervised (Random Forest) machine learning algorithms. The reconstruction of lost route segments is carried out using time-weighted linear interpolation. Testing the algorithms on telemetry data from 26 birds of prey across four species demonstrated the high efficacy of the hybrid approach during periods of active directional migration (the filtering efficiency of distorted locations averaged 98.7±2.2%). At the same time, the algorithms showed limitations when processing data from stationary areas (nesting, wintering, and prolonged stopovers), where anomaly recognition efficiency decreased significantly (to 69.6±45.2%).

The increasing volume and sampling frequency of telemetry data require a critical evaluation of algorithms for estimating animal home range (HR), especially for highly mobile species with complex territorial behavior. In this study, we conducted a comprehensive comparative analysis of the performance of 6 HR estimation algorithms using tracking data from Golden Eagles (Aquila chrysaetos). Model evaluation was performed using aggregated multi-criteria decision analysis (MCDA), spatial indices (SQI, M1/M2), predictive power metrics (AUC, LL), and the two-sample Kolmogorov-Smirnov test to assess behavioral realism. The analysis was conducted on both original high-frequency data and artificially rarefied (downsampled) tracks (up to 1 location per 6 hours). The results revealed fundamental differences in the spatial topology of the models. Classical approaches (KDE h ref, BBMM) demonstrated excessive oversmoothing of the HR core area (50%). In contrast, dynamic Brownian Bridge Movement Models (dBBMM) were highly sensitive to sampling density. On high-frequency data, the algorithm falls into spatial overfitting, classifying over 90% of actual transit locations as statistical noise. Extreme downsampling of tracks revealed a paradox of algorithm convergence and an inversion of their scale nesting. Also, it led to a critical loss of behavioral microstructure across all interpolative models. Based on the aggregation of all metrics, the AKDE method was recognized as the absolute leader. By integrating the temporal autocorrelation structure (OU/OUF models), AKDE proved to be the only algorithm that preserved the reliable spatial extent of the HR and predictive accuracy under sparse data conditions. This study demonstrates that for highly mobile avian predators, route interpolation becomes ecologically inadequate, and it is precisely AKDE that allows the transformation of discrete transit locations into a biologically robust HR model.

Force feedback drives sim racing performance: The influence of force and vibrotactile haptic feedback in simulator steering wheels

Improving our understanding of activity budgets in marine animals is critical to calibrating and refining bioenergetic models to predict larger-scale population trends and distributional shifts with climate change. Lab studies have been instrumental in identifying environmental influences on metabolism, particularly for commercially important species like snow crab ( Chionoecetes opilio ). However, extrapolating laboratory results to free-ranging organisms carries the risk of generating misleading interpretations, particularly for metabolic processes such as activity that emerge from multifaceted and context-specific mechanisms. We combined laboratory-derived energetic equations with telemetry data analyzed using Hidden Markov Models to estimate total in-situ standard and activity-related metabolic demand for snow crab living in Cabot Strait and the Grand Banks, regions that together encompass nearly the full range of the species' thermal distribution in Atlantic Canada. We compare these field estimates of metabolic demand to previously published studies to understand which components of metabolism are most important across sites and whether behavioral changes in activity compensate for increased metabolic costs at higher temperatures. Our study illustrates that thermal effects on standard metabolism is not a dominant driver of overall snow crab metabolism within the range of our observations, in contrast to lab studies. Instead, we observed active metabolism to be a more important contributor to total metabolism, and variation in movement patterns across study areas resulted in a moderating effect on lab-derived expectations of temperature on overall metabolism. • Hidden Markov Models were applied to better understand activity metabolism of free ranging snow crab. • Thermal effects on standard metabolism did not dominate wild snow crab metabolism across the broad thermal distribution studied. • Active metabolism was the most important contributor to total metabolism. • Thermal effects on realized activity metabolism were moderated by changes in movement patterns across study areas.

Miniature implantable neural interface devices are increasingly critical for both neuroscience research and clinical neuromodulation applications. However, device miniaturization imposes stringent constraints on power, area, and performance, creating challenges for implementing energy-efficient neuromodulation, high-fidelity neural recording, and wireless data telemetry. This review provides a comprehensive overview of low-power circuit designs enabling next-generation neural interfaces. We discuss energy-efficient stimulation drivers for optogenetic neuromodulation, highlighting advanced switched-capacitor-based techniques that reduce supply voltage requirements while maintaining high-current LED pulses. Low-noise neural recording frontends, including preamplifier-fronted structures, as well as ΔΣ ADC-based and NS-SAR-based direct-digitizing architectures, are reviewed with emphasis on techniques for dynamic range extension, linearity improvement, and artifact tolerance. Finally, state-of-the-art backscatter-based wireless telemetry methods are presented, covering load-shift keying (LSK), frequency-splitting, and push-pull quadrature modulation approaches that decouple power and data transfer to achieve high data rates with minimal energy consumption. This review highlights the critical role of circuit-level innovations in overcoming the power and performance limitations of miniature implants and provides insights for the design of next-generation neural interface systems.

The rapid proliferation of Internet of Things (IoT) devices and distributed computing platforms has accelerated the adoption of the edge–cloud continuum, an architectural paradigm that integrates edge devices, fog nodes, and centralized cloud infrastructures to support real-time data processing and latency-sensitive applications. While this architecture enhances scalability, responsiveness, and intelligent service delivery, it simultaneously expands the cyber-attack surface due to the presence of heterogeneous, resource-constrained, and geographically distributed devices. Traditional perimeter-based security mechanisms are increasingly inadequate for protecting such dynamic environments, while many existing Zero Trust Architecture (ZTA) implementations rely on static access control policies and centralized decision mechanisms that limit scalability and real-time responsiveness. This study proposes a Machine Learning-Driven Self-Healing Zero Trust Architecture (SH-ZTA) designed to enable autonomous cyber resilience across the edge–cloud continuum. The framework integrates Graph Neural Networks (GNNs) for relational anomaly detection and Deep Reinforcement Learning (DRL) for adaptive security policy orchestration. Network telemetry data collected from IoT devices and edge gateways are represented as communication graphs, enabling the detection of abnormal interactions, compromised nodes, and potential lateral movement attacks. The reinforcement learning agent dynamically enforces micro-segmentation policies, isolates malicious entities, and reconfigures network pathways to maintain operational continuity without human intervention. Experimental evaluation conducted in a simulated edge computing environment demonstrates that the proposed SH-ZTA framework significantly improves threat mitigation efficiency while maintaining low computational overhead suitable for resource-constrained devices. The results show improved detection accuracy, faster response latency, and enhanced network resilience compared to conventional security approaches.

The proliferation of Wireless Sensor Networks (WSNs) in mission-critical applications has made them primary targets for sophisticated routing layer threats, specifically multi-point wormhole attacks that compromise data integrity through artificial low-latency tunnels. This project proposes an Autonomous Self-Rerouting for Multi-Wormhole Mitigation in Wireless Sensor Networks using XGBoost Ensemble Learning to transition network security from passive detection to active, autonomous resilience. Initially, the framework ingests real-time telemetry data, including Round Trip Time (RTT) and Hop-Count Symmetry, which is refined using an Adaptive Feature-Aware Noise Suppression (AFNS) Logic to eliminate environmental jitter and synchronization artifacts. The refined data is then processed by an XGBoost-based Ensemble Classifier, which performs high-dimensional feature extraction to isolate the subtle signatures of colluding malicious nodes. To minimize false positives caused by natural network congestion, a Symptom-Aware Trust Engine (DTE) is integrated to evaluate node reliability over multiple transmission cycles. Once a threat is validated, an Autonomous Mitigation Layer is triggered to logically prune malicious edges from the network topology. The system then utilizes a Cost-Aware Dijkstra’s Algorithm to recalculate secure alternative paths in real-time, ensuring zero-downtime communication. Experimental results demonstrate that the proposed integrated approach maintains a Packet Delivery Ratio (PDR) above 95% even during intense attack scenarios. Ultimately, this framework provides a robust, self-healing solution that significantly improves the reliability and longevity of secure WSN infrastructures

Predictive maintenance for hydraulic excavators is difficult because component degradation is driven by interacting factors such as contamination, heat, mechanical imbalance, and leakage. This study develops an interaction-aware predictive maintenance framework that combines high-order hazard/survival modelling, gradient-boosted machine learning, and diagnostic visual analytics to predict time-to-service and support maintenance decisions. Telemetry and laboratory oil data were processed through integration, cleaning, feature engineering, and scaling. Decision Tree, Support Vector Regression, XGBoost, and LightGBM models were evaluated using k-fold cross-validation. The framework also incorporated principal component analysis, reliability assessment, calibration analysis, feature importance, and SHAP interaction surfaces. Risk maps, survival surfaces, and policy/cost planes were used to translate model outputs into operational actions. LightGBM achieved the best predictive performance for first-cycle maintenance and remained superior to baseline models in later cycles, despite reduced accuracy caused by cycle-dependent drift. Results consistently show oil cleanliness as the dominant maintenance driver, while temperature strongly amplifies failure risk. By capturing nonlinear interactions and changing degradation patterns across cycles, the proposed framework enables cycle-aware scheduling, improved contamination control, and thermal derating policies. The approach provides an interpretable and deployable toolkit for reducing unplanned downtime in heavy construction fleets.

Abstract Step‐selection functions (SSFs), typically fitted using step‐selection analysis (SSA) or integrated step‐selection analysis (iSSA) are widely used to infer habitat selection and movement kernels from high‐frequency telemetry data, but most standard validation tools focus on one‐step‐ahead prediction and do not guarantee that fitted models generate realistic trajectories or emergent space‐use patterns. We propose a multi‐criteria generative validation framework for SSF‐based movement models (typically fitted via SSA/iSSA), built around four pillars that target emergent utilization distributions, mean squared displacement, path sinuosity and barrier crossing. For each pillar, we define an ecologically interpretable trajectory‐level summary (Wasserstein distance between utilization distributions, mean squared displacement, straightness index and barrier‐crossing counts) and embed it in a Monte Carlo rank‐testing scheme that propagates parameter uncertainty. Applied to six synthetic ‘stress tests’ (sedentary home range, hard barrier, corridor follower, multi‐state movement, orbiter and return‐conditioned sinuosity), the framework reveals distinct failure modes that may not be detected by conventional stepwise validation. An empirical application to a GPS‐tracked red deer illustrates partial generative realism at a 6‐h sampling interval: the fitted iSSA reproduces long‐horizon space‐use and path sinuosity but shows a detectable mismatch in displacement dynamics (mean squared displacement).

Despite its continued growth, cloud computing remains susceptible to significant security challenges, as shared virtualised environments pose threats at multiple levels. These vulnerabilities are caused by a lack of security coverage in the responsibility model between the provider and the tenant. In this work, we propose the multi-layered architecture VIRTUOSO (VIRTual Unified Operation Security Optimiser) to cover these security gaps through advanced automation and ML. VIRTUOSO has four layers. The Input Layer extracts key risk components from collected telemetry data. The Deep Automation Security Layer provides automated actions and continuous monitoring of security defences. Its counterpart, the Intelligent Security Layer, predicts threats using anomaly detection. The last layer, the Output Layer, returns an aggregated risk summary. The datasets we used were chosen for their relevance: the UNSW-NB15 dataset, a subset of the web-attack classification from CSE-CIC-IDS2018, and a sample of anonymised log events from AWS CloudTrail. Our ensemble classifiers achieve a best accuracy of 95.08% ± 0.13% on UNSW-NB15 (RF), with statistically significant differences among models confirmed by the Friedman test (p < 0.004) and Nemenyi post hoc analysis, and 99.25% ± 0.52% on web-attack (CatBoost), where ensemble differences are not statistically significant (p = 0.093), consistent with the high separability of this dataset. The training-test gap and DNN curves show no overfitting, whereas our adversarial tests show a maximum accuracy loss of 8.1% at ε = 0.02. With these promising results, we can assert that, pending verification in an actual cloud environment and potential integration with FL, our ensemble classifier model appears to be a good real-world prototype.

Many migratory bird populations are declining in the face of habitat degradation and climate change, making it important to identify which stages of their annual cycle are most affected in order to guide conservation measures. The Bewick's Swan Cygnus columbianus bewickii , an Arctic‐breeding waterfowl species, has suffered a dramatic population decline (from approximately 30 000 individuals in 1995 to around 13 000 in 2020) due to low reproductive output. However, it is unknown which component of reproduction (breeding propensity, nesting success and post‐hatching survival) is the underlying cause. We analysed GPS tracking and accelerometer data from over 60 adult female Bewick's Swans collected over 8 years, to derive breeding propensity (i.e. the decision to breed or not) and nesting success (i.e. completing incubation) at the individual level, and assess how these metrics are influenced by onset of spring and migration timing. The odds of breeding increased by 3.9% (estimate = −0.04, se = 0.02, P = 0.04) for each day that the swans arrived earlier at the breeding grounds. The odds of successfully nesting increased by 5.8% (estimate = −0.06, se = 0.03, P = 0.06) per day of earlier snowmelt anomaly, suggesting a slight positive effect of climate warming on nesting. However, the proportion of breeding females was consistently low throughout the study period, suggesting that other factors may be driving the long‐term population decline of the species by negatively impacting breeding propensity. Our study shows that distinguishing between different phases of the reproductive cycle is critical for a better understanding of the causes of migratory population declines in a changing world.

Feral donkeys (Equus asinus) are well adapted to arid ecosystems and are found in large populations in the deserts of Australia and the Americas. We assessed resource selection and seasonal home range size of female donkeys in southern California between 2020 and 2022 based on telemetry data. We also examined whether dyads with greater encounter rates were more likely to test positive for asinine herpesvirus 5 (AHV-5) and/or Streptococcus equi zooepidemicus (SEZ). Donkey home ranges were non-significantly larger in the cool/wet season (November through March; mean 318.37 ± sd 417.65 km2) than in the hot/dry season (April through October; mean 159.35 ± 212.43 km2). Donkeys selected flatter areas closer to water year-round but selected greater herbaceous cover during the cool/wet season and lower heat loads during the hot/dry season. Individuals testing positive for SEZ selected lower elevations during the wet season and closer distances to water during the dry season; donkeys testing positive for AHV-5 selected areas farther from water during the wet season and steeper slopes during the dry season. The dyad encounter rate was unrelated to presence of either disease. Our results contribute to the understanding of donkey ecology, allowing feral populations to be better controlled by specific and focused management.

The authors discuss the concept of an active digital rod for high-precision leveling, based on using a display as a carrier of its digital code. The device is designed to overcome key limitations of passive barcode rods, namely the finite operating- and limited measurement ranges during the work, and pronounced dependence on external lighting conditions. The principle of the system`s functioning is outlined, involving dynamic formation of measurement patterns, using telemetry data, and arranging bidirectional information exchange with the leveling instrument. Potential advantages related to expanding the working range, increased accuracy in challenging conditions, and minimization of errors are analyzed, along with the systemic technological and economic barriers hindering practical implementation. It is concluded that, despite the complexity of development, this concept is of a particular interest for specialized tasks in the field of precision geodesy

Abstract: Cybersecurity threats have grown more complex and frequent, creating serious risks for organizations, critical infrastructure, and individuals worldwide. Traditional signature-based security tools can no longer effectively identify and deal with advanced, evasive, and quickly changing cyber-attacks, such as zero-day exploits, ransomware, and multi-stage intrusion campaigns. As a result, there is a strong need for real-time cyber threat detection and response systems that adjust dynamically and offer timely, actionable information to security operations teams. This paper provides a detailed review of modern methods that combine machine learning (ML) and open-source intelligence (OSINT) gathered through automated web data scraping. Machine learning offers powerful analysis for spotting both known and unknown threats by recognizing patterns and detecting anomalies in various telemetry data, including network traffic, system logs, and endpoint activities. OSINT enhances these systems by supplying external insights into new vulnerabilities, threat actor tactics, techniques, and procedures (TTPs), as well as real-time cyber threat intelligence shared across open channels like social media, security forums, paste sites, and the dark web[1][2][3].By combining ML-based internal monitoring with continuously updated OSINT feeds, advanced systems improve threat classification accuracy, lower false alarms, and provide contextual information that aids proactive responses. This review looks into the key architectures, machine learning algorithms, and natural language processing techniques for analyzing OSINT, along with illustrative case studies in IoT, finance, and healthcare. It also highlights existing challenges, such as managing data quality, ensuring model robustness, and addressing privacy and compliance issues. It outlines future research directions, focusing on federated learning, explainability, and blockchain-enabled threat intelligence sharing. This paper aims to be a valuable resource for researchers and practitioners seeking more effective, adaptable, and integrated cyber security defence frameworks that can tackle the increasingly sophisticated threat landscape.