Adel Binbusayyis

Collaborative cyber threat intelligence sharing is becoming a crucial part of contemporary security practices but is still limited by privacy risks and the need to retain confidence within and between organizations due to the heterogeneous nature of the information. NeuralFortress-XFL attempts to alleviate these issues by using a federated and explainable learning architecture that is capable of detecting threats with high fidelity, without revealing raw data or model explanations. The structure incorporates a Dynamic Privacy-Efficient Mechanism (DPEM), which dynamically adapts differential privacy budgets based on local data sensitivity and a lightweight explanation-embedding component that preserves both interpretability and privacy. An experiment conducted using CTI-Sharing-2024 dataset, which contains 15 organizations consisting of financial and healthcare industries, indicates that NeuralFortress-XFL provides the same level of detection as the centralized training accommodations, with significantly lower communication overhead. NeuralFortress-XFL maintains robust performance against up to 40% independent malicious participants (87.4% accuracy with label flipping, 84.3% backdoor detection rate). Against coordinated collusion, robustness degrades proportionally with coalition size: coalitions of five or more clients (33% of 15 organizations) achieve 67.3% attack success, a limitation that is explicitly addressed in the paper, and a trust-weighted aggregation process maintains global model consistency despite adversarial client updates. The framework is also able to offer consistent and sound explanations, which can support real-world analyst workflows. As a whole, NeuralFortress-XFL is a solid security solution for sharing privacy-sensitive cyber threat intelligence in a deployable and balanced way that satisfies all of the following criteria: it is accurate, it is interpretable, it communicates well, and it exhibits adversarial resilience in a multi-organizational real-world setting.

Deception, a widespread aspect of human behavior, has significant implications in fields like law enforcement, security, judicial proceedings, and social areas. Detecting deception accurately, especially in high-stakes environments, is critical for ensuring justice and security. Recently, machine learning has significantly enhanced deception detection capabilities by analyzing various behavioral and visual cues. However, machine learning models often operate as opaque "black boxes," offering high predictive accuracy without explaining the reasoning behind the decisions. This lack of transparency necessitates the integration of Explainable Artificial Intelligence to make the models’ decisions understandable and trustworthy. This study proposes the implementation of existing model-agnostic Explainable Artificial Intelligence techniques—Permutation Importance, Partial Dependence Plots, and SHapley Additive exPlanations—to showcase the contributions of visual features in deception detection. Using Real-Life Trial dataset, recognized as the most valuable high-stake dataset, we demonstrate that Multi-layer Perceptron achieved the highest accuracy of 88% and a recall of 92.86%. Along with the aforementioned existing techniques, Real-Life Trial dataset inspired us to develop a novel technique: ‘ set-of-features permutation importance’. Additionaly, this study is novel in the sense of that it extensively applies XAI techniques in the field of deception detection on Real-Life Trial dataset. Experimental results shows that the visual cues related to eyebrow movements are most indicative of deceptive behavior. Along with the new findings, our work underscores the importance of making machine learning models more transparent and explainable, thereby enhancing their utility for human-in-loop AI and ethical acceptability.

Security is considered as one of the primary characteristics in technical setup, as rise of IoT devices can be prone to various challenges, thus is extremely important to detect the attacks in order to prevent horrendous consequences, hence IDS (Intrusion Detection System) is utilized. Typically, different tactics are used for identification of attacks, however, manual approaches are considered to be time consuming and prone to error. Thus, AI based approaches are utilized for effective detection of attacks, however, prevailing works often lack in terms of accuracy and primarily focuses on binary classification of attacks. Hence, in order to overcome these pitfalls proposed framework utilizes proposed UN (Unify Net) and proposed NSSN (Neutro Sequential Sense Net) functions. Proposed UN model enhances the spatial features learning the property of LSTM model which amplifies the efficacy of the classifiers for intrusion detection. After employing proposed UN, proposed NSSN is primarily used for handling the issues associated to ambiguity information in intrusion data by extracting various deep features and helps in improving the performance and precision of model for multiclass classification and aids in preventing overfitting of the model. Dataset used in the model is IoT device network logs dataset, where five attacks and non-attack is specified. Eventually, the performance of the proposed mechanism is gauged by means of evaluation metrics. Besides, comparative analysis is carried out to compare the performance of the model and its ability to identify multiple attacks present in IoT network.

Prediction of stability in SG (Smart Grid) is essential in maintaining consistency and reliability of power supply in grid infrastructure. Analyzing the fluctuations in power generation and consumption patterns of smart cities assists in effectively managing continuous power supply in the grid. It also possesses a better impact on averting overloading and permitting effective energy storage. Even though many traditional techniques have predicted the consumption rate for preserving stability, enhancement is required in prediction measures with minimized loss. To overcome the complications in existing studies, this paper intends to predict stability from the smart grid stability prediction dataset using machine learning algorithms. To accomplish this, pre-processing is performed initially to handle missing values since it develops biased models when missing values are mishandled and performs feature scaling to normalize independent data features. Then, the pre-processed data are taken for training and testing. Following that, the regression process is performed using Modified PSO (Particle Swarm Optimization) optimized XGBoost Technique with dynamic inertia weight update, which analyses variables like gamma(G), reaction time (tau1–tau4), and power balance (p1–p4) for providing effective future stability in SG. Since PSO attains optimal solution by adjusting position through dynamic inertial weights, it is integrated with XGBoost due to its scalability and faster computational speed characteristics. The hyperparameters of XGBoost are fine-tuned in the training process for achieving promising outcomes on prediction. Regression results are measured through evaluation metrics such as MSE (Mean Square Error) of 0.011312781, MAE (Mean Absolute Error) of 0.008596322, and RMSE (Root Mean Square Error) of 0.010636156 and MAPE (Mean Absolute Percentage Error) value of 0.0052 which determine the efficacy of the system.

In the modern world, the evolution of the internet supports the automation of several tasks, such as communication, education, sports, etc. Conversely, it is prone to several types of attacks that disturb data transfer in the network. Efficient attack detection is needed to avoid the consequences of an attack. Traditionally, manual attack detection is limited by human error, less efficiency, and a time-consuming mechanism. To address the problem, a large number of existing methods focus on several techniques for better efficacy in attack detection. However, improvement is needed in significant factors such as accuracy, handling larger data, over-fitting versus fitting, etc. To tackle this issue, the proposed system utilized a Random Grove Blend in Weighted MLP (Multi-Layer Perceptron) Layers to classify network attacks. The MLP is used for its advantages in solving complex non-linear problems, larger datasets, and high accuracy. Conversely, it is limited by computation and requirements for a great deal of labeled training data. To resolve the issue, a random info grove blend and weight weave layer are incorporated into the MLP mechanism. To attain this, the UNSW–NB15 dataset, which comprises nine types of network attack, is utilized to detect attacks. Moreover, the Scapy tool (2.4.3) is utilized to generate a real-time dataset for classifying types of attack. The efficiency of the presented mechanism is calculated with performance metrics. Furthermore, internal and external comparisons are processed in the respective research to reveal the system’s better efficiency. The proposed model utilizing the advantages of Random Grove Blend in Weighted MLP attained an accuracy of 98%. Correspondingly, the presented system is intended to contribute to the research associated with enhancing network security.