Detecting and eliminating fake profiles is crucial to maintaining the integrity of social media platforms and ensuring user safety. This paper proposed a hybrid artificial intelligence model to address these challenges by combining advanced preprocessing, feature extraction and classification techniques. For feature extraction, principal component analysis (PCA) is used to decrease the dimensionality of the data, ensuring efficient processing. The hybrid classification framework integrates a 2D convolutional neural network (2D CNN) to extract spatial features from the input data. Additionally, an attention-enhanced stacked BiLSTM (AES-BiLSTM) is utilised to capture temporal features, while the added attention mechanisms improve the model's focus on the most relevant information. The 2D CNN and AES-BiLSTM models are further enhanced through hybrid optimisation, with their hyperparameters fine-tuned using the seagull optimisation algorithm (SOA). This approach achieves a high accuracy of 98.9%, precision of 98.9%, specificity of 98.9%, and an F1-score of 99%.
Malicious nodes pose growing threats to mobile networks by disrupting communication and reducing performance, while existing detection methods often suffer from high overhead and poor scalability. To address this, this study proposes a chronological addax optimisation algorithm based siamese convolutional neural network (CrAOA-SCNN), CrAOA combined with a Siamese CNN for malicious node detection. The proposed framework begins with a mobile network simulation, where routing is managed using the ad hoc on-demand distance vector (AODV) protocol and node trustworthiness is estimated through a trust evaluation mechanism. The SCNN model identifies malicious nodes, while its hyperparameters are optimised using CrAOA. The proposed routing strategy results in a delay of 0.247 sec, a data packet delivery ratio of 92.89%, and a trust of 88.89%, as well as the superior accuracy, true negative rate (TNR), and true positive rate (TPR) of 91.37%, 90.84%, and 90.49% are achieved in malicious node detection.
Although large-scale cyberattacks on wearable medical devices (WMDs) are not yet widespread, the growing complexity and connectivity of these systems make them a potential target for battery-draining threats. Devices such as pacemakers, insulin pumps, and neurostimulators are critically dependent on a stable power supply to perform life-sustaining functions. This article explores the possible causes of increased energy consumption, including cyber impacts on sensor, computational, and communication modules. In response to these challenges, the paper proposes an approach that combines adaptive power management with a lightweight intrusion detection system tailored to the specific characteristics of WMDs. The study emphasises the importance of implementing energy-efficient and proactive strategies to enhance the resilience of these devices against emerging cyber-physical threats.
Cyberattacks are becoming increasingly sophisticated, making it challenging for network intrusion detection systems (NIDS) to detect rare attacks due to severe class imbalance in network traffic data. This study evaluates LightGBM, XGBoost, and random forest models combined with data balancing techniques (ADASYN, SMOTE, and random oversampling) using the UNSW-NB15 and CICIDS2017 datasets. Results demonstrate that oversampling methods enhance minority class detection, with ADASYN achieving the highest performance - improving recall for 'Worms' from 0.528 to 1.000 and F1-score for 'Shellcode' from 0.834 to 1.000. On CICIDS2017, the model reached 99.96% accuracy, even for rare attacks like Heartbleed. Key innovations include clustering-based data reduction, outlier removal via IQR, and comparative analysis of augmentation methods. The study confirms ADASYN's effectiveness in addressing imbalance and boosting NIDS sensitivity, offering a practical solution for real-world cybersecurity applications.
Online social network automation attacks (OSNAA) increasingly employ automated tools to perform malicious activities such as bot-based interactions, email hijacking, and malware-driven manipulation. This study introduces an automated social network attack detection model (ASNADM) that integrates energy consumption footprint analysis (EComp-FP) with automated software opcode sequence analysis (ASOSA-OSM) to identify automation-based anomalies at the client side. The framework analyses behavioral energy traces from CPU and system activity alongside opcode n-gram representations derived from executable binaries using opcode frequency variance (OFV) and weighted term frequency (TF-W). These heterogeneous features are fused and classified using self-adaptive soft fuzzy C-means (SSFCM) clustering. Experiments conducted on the SPEMC-15K-E dataset demonstrate a detection accuracy of 99.93% with a 0.07% false-positive rate, outperforming DT, KNN, RF, and SVM models. Results confirm that abnormal energy patterns and low-entropy opcode sequences effectively reveal malicious automation in online social networks.
