Muhammad Azfar Yaqub

Accurate trajectory prediction and proactive Road-Side Unit (RSU) handover are essential for maintaining seamless connectivity and low-latency Multi-Edge Computing (MEC) based services in Cooperative Adaptive Cruise Control (CACC) systems. Conventionally, mobility forecasting and migration decisions are treated as separate problems, neglecting their inherent dependence, thus leading to premature or delayed handovers. In this work an integrated learning-based pipeline is introduced that first predicts the vehicle mobility using deep sequential models and the leverages these predictions to trigger intelligent RSU migrations. We employ a LSTM network to learn temporal dynamics from the large-scale FHWA dataset and generate multi-step displacement predictions over short horizons. These predictions are then used to infer future coverage boundaries and connectivity risks. Further, we model RSU migration as a binary-class classification task and train Random Forest (RF) and Multilayer Perceptron (MLP) as classifiers on engineered mobility features, such as velocities deltas, cumulative drift and distance to RSU markers. To handle skewed migration labels, we incorporate fallback heuristics and class-balanced strategies. Experiments on the 500,000 real vehicular samples reveal that the RF-based migration model achieves up to 73% accuracy. Meanwhile LSTM maintains stable short-horizon displacement accuracy with competitive Average Displacement Error (ADE) and Final Displacement Error (FDE) scores. Together, the models enable anticipatory RSU handover decisions that outperform naive threshold-based methods.

Developing data-driven models for spatiotemporal hydrological prediction presents challenges in managing complexity, capturing fine spatial and temporal resolution, and ensuring model resilience across diverse regions. This study introduces an innovative surrogate deep learning (SDL) architecture designed to predict daily soil moisture (DSM) and daily actual evapotranspiration (DAE) by integrating climate data and geophysical insights, with a focus on mountainous areas such as the Adige catchment. The proposed framework aims to enhance the parameter-calibration quality. The process begins by mapping the statistical characteristics of DAE and DSM across the whole region using an unsupervised fusion technique. Model accuracy is assessed by comparing the similarity of Fuzzy C-Means (FCM) clusters before and after fusion, providing a metric for feature reduction. A data transformation technique using Gradient Boosting Regression (GBR) is then applied to each homogeneous subregion identified by the Random Forest classifier (RFC), based on elevation parameters (Wflow_dem). Furthermore, Kernel density estimation is used to ensure the reproducibility of the RFC-GBR process across large-scale applications. A comparative analysis is conducted across multiple SDL architectures, including LSTM, GRU, TCN, and ConvLSTM, over 50 epochs to better evaluate the beneficial effect of the transformed parameters on model performance and accuracy. Results indicate that adjusted parameter calibration improves model performance in all cases, with better alignment to Wflow ground truth during both wet and dry periods. The proposed model increases the accuracy by 20% to 42% when using simpler SDL models like LSTM and GRU, even with fewer epochs.

The globalization and digitization of society have caused a surge in network traffic, making reliable online services essential for user trust and system functionality. However, these services face ever-increasing threats, particularly complex and well-developed Distributed Denial of Service (DDoS) attacks. Zero-day DDoS attacks, a type of DDoS attack, are especially challenging because their new and unseen nature and lack of training data render traditional Intrusion Detection and Prevention Systems (IDPS) ineffective. To tackle this, we propose the Fog-based One Solution For All (F-OSFA) system - a model with three specialized components. The first component uses a hybrid machine learning and deep learning framework that combines convolutional neural networks (CNNs) and decision trees to detect traditional DDoS attacks. The second component employs a few-shot learning module with a contractive autoencoder for zero-day attack detection. The third component is a signature-based resource usage analyzer to counter attacks mimicking normal traffic. We demonstrate the efficacy of F-OSFA on publicly available datasets and prove the scheme is generalizable and effective. F-OSFA achieves an accuracy of 99.72% on CICDDoS2019 and 99.96% on CICIDS2017. In addition, it demonstrates its efficacy in the zero-day scenario as well by achieving a 96.77% on CICDDoS2019 and 95.98% on CICIDS2017. These evaluations testify to F-OSFA as a reliable and versatile solution against ever-evolving DDoS threats.