Muhammad Zeeshan Shakir

Rainfall prediction models often lose skill when transferred across regions, particularly in data-sparse settings where local recalibration is not feasible. This study investigates whether topographically analogous landscapes exhibit consistent patterns of model portability in next-day rainfall occurrence prediction across contrasting climates. Random Forest and Logistic Regression classifiers trained on multi-year daily atmospheric data were evaluated across three hydrogeomorphic classes, Alluvial/Valley, Delta/Marsh, and Coastal Plain, using paired temperate sites in the United States and tropical sites in Malaysia. Portability was quantified using Transferability, representing a model’s ability to export predictive skill, and Adaptability, representing a site’s receptivity to externally trained models. Performance, sensitivity, and stability metrics were synthesized within a Behaviour Grid to support systematic interpretation. Results reveal a robust terrain-driven hierarchy: Delta/Marsh models are the strongest exporters, Alluvial/Valley sites the most adaptable receivers, and Coastal Plain sites are stable generalists. This general hierarchy is preserved across climates and years when evaluated using Area Under the Receiver Operating Characteristics Curve, indicating strong terrain structuring of portability, although fixed-threshold performance metrics show sensitivity to temporal variability. These findings indicate that terrain class provides a useful basis for evaluating cross-regional model transfer, supporting more informed regionalisation and deployment decisions across heterogeneous environmental settings.

As extreme heat events increase in frequency, intensity, and duration due to climate change, forecasting these events has become vital for early warning systems, public health preparedness, and climate adaptation strategies, especially in parts of the world that are already subject to extreme heat, such as tropical regions. In recent years, machine learning (ML) has increasingly been applied to environmental and meteorological data to improve the prediction of heatwaves and extreme heat events. This scoping review examines global peer-reviewed literature on the application of ML techniques for extreme heat prediction using environmental variables. This includes heatwave prediction, environmental and meteorological predictors used in these models, and the geographical distribution of existing research. A total of 23 peer-reviewed studies meeting the inclusion criteria were included in the review, following the PRISMA-ScR guidelines. The findings indicate that artificial neural networks and random forest models were most frequently reported as high performing within individual studies. However, direct comparisons across studies are limited by heterogeneity in prediction targets, validation strategies, lead times, heatwave definitions, and performance metrics. Temperature-related variables, especially maximum temperature, were consistently identified as the most influential predictors across studies. Furthermore, the evidence base was heavily concentrated in Europe, Asia, and North America, with comparatively limited representation from low- and middle-income countries respective to population, despite these regions often experiencing disproportionate impacts of climate change and extreme heat exposure. By synthesising current evidence on ML-based heatwave prediction, associated environmental predictors, and geographical research trends, this review provides insights to support the development of more robust, context-aware, and globally representative heatwave forecasting frameworks.

As extreme heat events increase in frequency, intensity, and duration due to climate change, forecasting these events has become vital for early warning systems, public health preparedness, and climate adaptation strategies, especially in the hottest parts of the world. In recent years, machine learning (ML) has increasingly been applied to environmental and meteorological data to improve the prediction of heatwaves and extreme heat events. This scoping review examines global peer-reviewed literature on the application of ML techniques for extreme heat prediction using environmental variables. This includes heatwave prediction, environmental and meteorological predictors used in these models, and the geographical distribution of existing research. A total of 23 peer-reviewed studies meeting the inclusion criteria were included in the review, following the PRISMA-ScR guidelines. The findings indicate that artificial neural networks and random forest models are among the most frequently reported high-performing approaches for heatwave prediction. Temperature-related variables, especially maximum temperatures, were consistently identified as the most influential predictors across studies. Furthermore, the evidence base was heavily concentrated in Europe and North America, with comparatively limited representation from low- and middle-income countries, despite these regions often experiencing disproportionate impacts of climate change and extreme heat exposure. By synthesising current evidence on ML-based heatwave prediction, associated environmental predictors, and geographical research trends, this review provides insights to support the development of more robust, context-aware, and globally representative heatwave forecasting frameworks.

Fifth-generation (5G) mobile networks are widely positioned as key enablers of industrial digital transformation. However, despite extensive coverage expansion, the deployment landscape remains dominated by Non-Standalone (NSA) architectures integrated with legacy 4G cores, limiting the practical availability of advanced capabilities such as Ultra-Reliable Low-Latency Communication (URLLC), Massive Machine-Type Communication (mMTC), and network slicing. This has contributed to a disparity between projected 5G functionality and realised industrial utility. This paper investigates the economic and structural factors constraining advanced 5G adoption and examines their implications for emerging sixth-generation (6G) frameworks. We conceptualise the current stagnation as arising from concurrent supply-side and demand-side constraints: elevated Radio Access Network (RAN) capital expenditure relative to previous generations, and limited demonstrable return on investment (ROI) for advanced service capabilities. To evaluate these dynamics empirically, a regional stakeholder study was conducted across industrial and public sector organisations in Ayrshire, Scotland. Data were collected through structured surveys and workshop-based questionnaires involving 34 participants, with proportional sectoral analysis performed to assess representativeness. The results indicate that high initial deployment costs and ROI uncertainty are the primary adoption barriers, with 45.83% of respondents reporting no immediate operational requirement for advanced 5G features. The findings identify an implementation gap in which economic viability, rather than technical feasibility, limits progression beyond basic 5G deployment. The paper argues that unless cost-efficiency and sector-specific value articulation are addressed, similar adoption constraints may extend into 6G development. These results provide empirically grounded insights to inform more economically aligned next-generation network planning.

The UK telecommunications sector’s 5G rollout is projected to consume 2.1% of national electricity by 2030, raising urgent sustainability concerns. This study empirically investigates, under controlled laboratory conditions, the energy performance and cost characteristics of two private 5G architectures—Vodafone’s Mobile Private Network (MPN) and an Open Radio Access Network (O-RAN) via BubbleRAN—and contextualises them against public network references and the United Nations Sustainable Development Goals (SDGs). Two complementary dimensions of energy performance are assessed: absolute power consumption (Watts), reflecting total system draw regardless of throughput; and throughput efficiency (Mbps/W), capturing useful data delivered per unit of energy. In terms of absolute power, O-RAN consumes less (460 W active, 378 W idle) than MPN (645 W active, 620 W idle). In terms of throughput efficiency, MPN delivers 1.45 Mbps/W versus O-RAN’s 0.44 Mbps/W under these specific controlled, single-cell conditions, a difference that reflects the tested hardware configurations (n77 vs. n78 band; 936 Mbps vs. 202 Mbps throughput; 2 × 2 vs. 4 × 4 MIMO) as much as any intrinsic architectural distinction. Both architectures offer substantially lower annual energy costs (£1060–£1486) compared to public micro-cells (£1991–£2666), representing 44–60% savings. Session continuity was 100% across all controlled trials; this reflects short-term laboratory conditions and should not be extrapolated to a long-term network availability guarantee without extended field validation. These results are configuration-specific preliminary indicators; the relative efficiency advantage of each architecture is expected to vary with load, band, and deployment scale. By 2030, UK 5G network operations are projected to generate 795,347–1,260,532 tonnes of CO₂ annually across low-to-high demand scenarios; private deployment, by reducing site proliferation 15–33%, could displace a meaningful share of this footprint. These findings support SDGs 4, 8, 9, 12, and 13. Hybrid O-RAN–MPN pilots are recommended to maximise sustainability gains while advancing social equity and net-zero targets.

Potato diseases pose a significant threat to global food security, making accurate automated detection systems essential for sustainable agriculture. The study presents a rigorous comparative evaluation of four deep learning architectures, namely EfficientNet‐B0, vision transformer (ViT‐B/16), ResNet‐50, and a U‐Net encoder, for classifying seven potato diseases plus healthy specimens using a dataset of 2529 images characterised by substantial class imbalance, with the two blight classes comprising 79% of samples. Several critical methodological gaps were addressed in the experimental design. Dataset splitting was performed before augmentation to prevent data leakage, stratified five‐fold cross‐validation was used to produce robust performance estimates, and paired statistical testing was applied to validate claimed model superiority. The augmentation strategy deliberately excluded texture‐destroying transformations such as Gaussian blur; ablation studies confirmed that including blur reduced accuracy by 4.65% points by degrading lesion boundary detail and fungal surface markings essential for disease discrimination. EfficientNet‐B0, ResNet‐50, and ViT‐B/16 were initialised from ImageNet pre‐trained weights, while U‐Net was trained from scratch owing to the absence of standard pre‐trained weights for that architecture. EfficientNet‐B0 achieved the highest overall performance, reaching 98.43% accuracy (±0.41%) with 92.37% macro‐averaged precision. Despite requiring 16.2 times fewer parameters than ViT‐B/16 (5.3 million versus 86.0 million), EfficientNet demonstrated substantially faster inference (12.3 ms versus 45.2 ms) and lower memory consumption (2.1 GB versus 7.8 GB). On the most challenging minority class, Blackleg, EfficientNet achieved 78% precision compared with ViT’s 67%, ResNet‐50’s 55%, and U‐Net’s 30%. ResNet‐50 achieved 96.85% overall accuracy but struggled on under‐represented classes, whilst U‐Net encoder produced the weakest results (91.75% accuracy, 57.55% macro‐precision), with near‐random performance on several minority classes, illustrating the critical role of transfer learning on small datasets. Statistical testing confirmed EfficientNet’s superiority over ResNet‐50 and U‐Net ( p < 0.01), while its advantage over ViT approached but failed to reach significance at the corrected threshold ( p = 0.051). In order to evaluate the robustness of these findings beyond controlled laboratory conditions, the four architectures were subsequently retrained and evaluated on a curated real‐world dataset combining the PotatoCare benchmark with the Mendeley ‘Novel Dataset of Potato Leaf Disease in Uncontrolled Environment’. EfficientNet‐B0 maintained its position as the highest‐performing architecture under these more demanding conditions, achieving 96.07% cross‐validation accuracy and 91% test‐set accuracy, confirming that the rankings established under laboratory conditions are preserved when variability in image quality and background content is introduced. The work establishes a rigorous algorithmic baseline and demonstrates that EfficientNet‐B0 represents the optimal choice for resource‐constrained agricultural deployment, balancing superior classification accuracy with exceptional computational efficiency.

In recent times, unmanned aerial vehicles (UAVs) have been used in many real-world applications such as trans-portations and delivery, agriculture and forestry, infrastructure inspection and surveying, surveillance, and counter-terrorism, wireless communications, and emergency and disaster management. For their role in disaster prevention and control, UAV operations as flying base stations (FBSs) have become pivotal in enhancing network coverage and quality of service. Therefore, autonomous FBS deployment has gained significant research interest due to UAV control and manoeuvring in dynamic and challenging environments. Further, these deployments hinge on efficient trajectory planning along with obstacle avoidance to reach their destination without depleting limited battery resources. To tackle this challenge, wireless channel characteristics such as channel state information (CSI) can help UAVs avoid obstacles and reach their destination point. In this work, we propose a novel deep learning-based FBS trajectory prediction framework that uses CSI information to dynamically learn environmental characteristics while avoiding obstacles and minimising flight time. By integrating a recurrent convolutional neural network (RCNN) architecture, our framework achieves a remarkable 98.34% prediction accuracy in identifying optimal trajectories, significantly outperforming deep reinforcement learning (DRL)-based 85.42% and traditional machine learning (ML)-based schemes 93.25%. Extensive simulations demonstrate that our approach reduces the flight time of the FBS by nearly 44 seconds (47.4%) compared to existing methods under obstacle-based conditions. Our approach demonstrates energy-efficient FBS trajectory planning and underscores its significance for applications in wireless communications, emergency response, and disaster management.

Despite the widespread adoption of network monitoring tools, Managed Service Providers (MSPs), specifically small- and medium-sized enterprises (SMEs), continue to face persistent challenges in achieving predictive, multi-tenant-aware visibility across distributed client networks. Existing monitoring systems lack integrated predictive analytics and edge intelligence. To address this, we propose an AI- and IoT-driven monitoring and visualisation framework that integrates edge IoT nodes (Raspberry Pi Prometheus modules) with machine learning models to enable predictive anomaly detection, proactive alerting, and reduced downtime. This system leverages Prometheus, Grafana, and Mimir for data collection, visualisation, and long-term storage, while incorporating Simple Linear Regression (SLR), K-Means clustering, and Long Short-Term Memory (LSTM) models for anomaly prediction and fault classification. These AI modules are containerised and deployed at the edge or centrally, depending on tenant topology, with predicted risk metrics seamlessly integrated back into Prometheus. A one-month deployment across five MSP clients (500 nodes) demonstrated significant operational benefits, including a 95% reduction in downtime and a 90% reduction in incident resolution time relative to historical baselines. The system ensures secure tenant isolation via VPN tunnels and token-based authentication, while providing GDPR-compliant data handling. Unlike prior monitoring platforms, this work introduces a fully edge-embedded AI inference pipeline, validated through live deployment and operational feedback.