Earthquakes are among the most hazardous natural disasters, posing significant threats to infrastructure, property and human life. This is primarily due to the sudden nature of earthquakes, which often provide little to no time for preparation. Consequently, the issue of earthquake prediction is crucial for human safety. Developing a reliable and highly accurate earthquake prediction model using machine learning (ML) methods can enhance our understanding of these complex natural phenomena, ultimately aiding in preserving lives and mitigating earthquake-related damage. In this study, we propose a novel feature selection approach that integrates two methods: normalisation based on analysis of variance and the Chi-squared technique, along with correlation based on Logistic Regression (CLR-AVCH). This approach aims to identify the most relevant features to expedite model training, minimise errors and optimise outcomes. We employ three algorithms (Support Vector Machine, Decision Tree and Random Forest) to uncover and identify patterns in the collected data. A soft voting classifier is then constructed, combining the best-performing models (Decision Tree and Random Forest) to create a unified model that leverages both strengths, improving prediction accuracy. The proposed methodology achieves high-performance metrics, including accuracy, F1 score, recall and precision (0.99, 0.98, 0.98 and 0.98, respectively). Future work will focus on implementing new feature selection techniques alongside hybrid algorithms with soft voting classifiers to enhance diagnostic capabilities.
The propagation of Internet of Things (IoT) devices in healthcare, while remote monitoring has enabled and patient care has improved, significant vulnerabilities have introduced that threaten the security of sensitive medical data. To address this crucial challenge, the study proposes a novel framework to enhance the privacy of data and trustworthiness in healthcare IoT ecosystems using advanced anomaly detection. We introduced a Customized Sand Cat Swarm driven Updated Random Forest (CSCS-URF) model, as a first step, data is preprocessed using min-max normalization and Recursive Feature Elimination (RFE). The Second step is analyzing data by an ensemble classifier optimized via a customized swarm intelligence algorithm to effectively identify security anomalies. Then, evaluate the system on the CICIoT20232 dataset, the proposed CSCS-URF method shows superior performance compared to existing benchmarks, achieving an accuracy of 97%, precision of 94%, recall of 96%, and an F1-score of 95%. These results indicate that the CSCS-URF framework is a robust solution for proactive security, enabling the early detection of potential breaches to strengthen system integrity and safeguard patient data against evolving cyber threats in the healthcare sector.
The Internet of Things (IoT) has enabled the rapid development of Cyber Physical Systems (CPS), but this has also created serious security concerns, particularly in the area of identifying sophisticated and changing cyberattacks. Traditional intrusion detection systems (IDSs) are built on static models and handcrafted features that are unable to keep up with the constantly changing nature of network behavior and new threats. This study presents a hybrid intelligent architecture built on Graph Neural Networks (GNNs) and Deep Reinforcement Learning (DRL) for context-aware and adaptive intrusion detection in order to address these limitations. In order to aid the GNN component in capturing complex spatial and relational patterns between connected nodes over time, the suggested method treats the network traffic as a dynamic graph made up of devices as nodes and interactions between them as edges. With a multi-objective reward that takes into account detection performance, false positive rate, and computation latency, a DRL agent is also presented as a meta-controller that dynamically adjusts detection policies in response to the current network states and environmental feedback. The framework is evaluated using three cutting-edge IoT intrusion detection datasets Edge-IIoTset, TON_IoT, and CICIoT2023 under a variety of traffic scenarios, including concept drift and zero-day exploits. The suggested method achieves a 97. 2% F1 score and a 1. 1% false positive rate, according to experimental data, which is significantly better than conventional machine learning, deep learning, GNN-only, and DRL-only baselines. When subjected to unidentified assaults, the end-to-end framework only experiences a 6. 2% drop in the F1-score, compared to 18. 3% for static GNNs, and it recovers from concept drift 4. 7 times faster than DRL-only techniques, all while running with a low inference latency (27. 6 ms per batch). These findings show how promising adaptive decision-making and integrated structural graph learning are for safeguarding the next generation of IoT-CPS systems.
The rapid advancement of modern deep neural networks (DNNs) has played a crucial role in aiding humans across many real-world applications; yet, their hardware accelerators have been proven to be vulnerable to malicious attacks. One particularly severe and serious attack involves inserting a hardware Trojan (HT) into DNN accelerator hardware in order to enable attackers to stealthily manipulate model predictions during the supply chain. In this paper, we present a possible stealthy HT architecture that is difficult to detect and has a significant impact on the performance of DNN models. To successfully achieve this goal, we introduce the Sensitivity-Based Weight Selection (SBWS) algorithm, a novel technique that adapts machine learning (ML) sensitivity analysis to identify and modify a small number of weights that have the highest impact on DNN performance, compared to previous work. We evaluate the proposed attack on five DNN model tests (two distinct DNN models and four different datasets) using two designed payload types (weight zeroing and sign-flipping) and record the results based on various security metrics. The experimental results show average accuracy reductions of 26.7% for the zeroing attack and 48.1% for the sign-flipping attack, yielding an overall average of 37.4%, calculated over five independent runs per dataset with standard deviation <2%. The sign-flipping technique consistently outperforms zeroing because it preserves the magnitudes of the attacked weights while inverting their signs, thereby disrupting the learned decision boundaries more severely and amplifying error propagation in subsequent layers. These results significantly exceed those of previous random-weight perturbation attacks (typically 12–20% drops) and other targeted HT approaches while incurring lower computational and hardware resource overheads. This work provides a more effective and scalable method for assessing the vulnerability of DNN accelerators under real supply chain threat models.
Multi-agent reinforcement learning (MARL) has been increasingly utilized for the real-time control of smart energy infrastructures, where agents are still vulnerable to false data injection attacks (FDIAs), Byzantine failures, and specification misalignment. Existing constitutional AI techniques target discrete language-based coordination, not continuous-action cyber–physical control. In this paper, we propose SafeAgentGrid, a hierarchical neuro-symbolic constitutional framework. A projection filter guarantees every hard box-type constraint by construction. A hybrid neural-surrogate mediator combines a low-latency fast path, admitted only after a split-conformal test bounds its alignment shortfall, with a full path over a tightenedDC reachable set under fairness (Gini) and risk-aware security weighting. Finally, 8-bit quantized models support edge computing. In a data-informed DC power-flow simulation, SafeAgentGrid accomplishes an average reward of 200,186.9 at violations/step, rising only to under 20% FDIA and under Byzantine failure of agents, at 100% grid efficiency and 0.79 ms/step mediation latency. It enhances both reward and safety over four unmediated techniques trained in the same environment (Wilcoxon signed-rank, ) on all ten training seeds and outperforms a joint safety shield and a Lagrangian constrained MARL technique, which violate the limits on 57.5% and 98.8% of the steps. We further derive the tightening from a chance constraint and show that it is necessary under the stochastic realization.
The rapid advancement of modern Deep Neural Networks (DNNs) has played a crucial role in aiding humans in many real-world applications, yet their hardware accelerators have been proven to be vulnerable to malicious attacks. One particularly severe and serious attack involves inserting a hardware Trojan (HT) into DNN accelerator hardware in order to enable attackers to stealthily manipulate model predictions during the supply chain. In this paper, we present a stealthy HT that is difficult to be detected and has a significant impact on the performance of DNN models. To successfully achieve this goal, we introduce the Sensitivity-Based Weight Selection (SBWS) algorithm, a novel technique that adapts machine learning (ML) sensitivity analysis to identify and modify only a small number of weights that have the highest impact on DNN performance, compared to previous work. We evaluate our proposed attack on five DNN models and multiple datasets using two designed payload types: weight zeroing and sign-flipping, and record the results based on various security metrics. The experimental results show average accuracy reductions of 26.7% for the zeroing attack and 48.1% for the sign-flipping attack, yielding an overall average of 37.4%, calculated over five independent runs per dataset with standard deviation < 2%. The sign-flipping technique consistently outperforms zeroing one because it preserves the magnitudes of the attacked weights while inverting their signs, thereby disrupting the learned decision boundaries more severely and amplifying error propagation in subsequent layers. These results significantly exceed previous random-weight perturbation attacks (typically 12–20% drops) and other targeted HT approaches, while incurring lower computational and hardware resource overheads. This work provides a more effective and scalable method for assessing the vulnerability of DNN accelerators under real supply-chain threat models.