Explaining AI Decisions to Mitigate Cognitive Biases in Human-AI Collaboration

Aiswan; Citra

doi:https://doi.org/10.54216/JCHCI.110205

International Journal of Wireless and Ad Hoc Communication | 2026

Communication-A ware Digital-Twin Reliability Budgeting for Fog-Assisted Wireless Sensor Ad Hoc Networks

Wireless sensor IoT systems are increasingly deployed as infrastructure-light communication fabrics in which battery-powered devices exchange event streams through local gateways, fog nodes, and sometimes multi-hop ad hoc routes. In such settings, reliability cannot be judged only by how fast a packet reaches a server. A reading may be fresh but untrusted, energy-efficient but delayed, or successfully delivered through a route that overloads the next fog node. This article revises the problem as a communication-aware reliability budgeting task for fog assisted wireless sensor ad hoc networks. It reviews core studies on wireless sensor networking, fog and edge computing, digital twins, edge intelligence, federated learning, and IoT security, then introduces an extended Digital-Twin Reliability Budgeting model. The model maintains compact fog-side twin states and uses them to govern route choice, event compression, fog offloading, replication, and cloud escalation. Three mathematical algorithms are presented for twin synchronization, route-and-action selection, and adaptive budget learning. The analysis develops delay, energy, freshness, loss, trust, and occupancy terms and shows how they interact across multi-hop communication paths. The resulting framework supports a more disciplined design philosophy: fog nodes should not only process sensor data near the edge; they should regulate the reliability budget of each communication decision before network resources are consumed.

groups

Safina Shokeen mail -

Vishal Srivastava mail

link https://doi.org/10.54216/IJWAC.100106

Volume & Issue

Vol. Volume 10 / Iss. Issue 1

Details open_in_new

International Journal of Wireless and Ad Hoc Communication | 2026

Edge-Bandwidth Brokerage for RFID-Enabled Ad Hoc IoT Communication Networks

RFID-enabled IoT deployments often lose communication efficiency not because a single tag message is large, but because tag observations are repeated, partially redundant, and reported by overlapping readers. In an edgecomputing environment, this redundancy becomes a bandwidth-governance problem: the local gateway must decide what is worth forwarding, what can be compressed, and what should be kept only as short-term local evidence. This article presents an edge-bandwidth brokerage model for RFID-assisted ad hoc IoT communication networks. The proposed model, named BASER, interprets every tag read as a priced communication event whose forwarding value depends on novelty, duplication risk, priority, motion context, and instantaneous backhaul pressure. The paper develops a three-stage mathematical formulation for value construction, budgeted admission, and adaptive compression. A reproducible scenario analysis is used to study how tag density, mobility, edge load, and uplink capacity affect latency, loss, semantic retention, and energy consumption. The main finding is that bandwidth savings should not be treated as a blind compression target; instead, the edge node should act as a broker that protects meaningful RFID events while preventing repeated low-value reads from saturating the uplink.

groups

Salah-ddine KRIT mail

link https://doi.org/10.54216/IJWAC.100207

Volume & Issue

Vol. Volume 10 / Iss. Issue 2

Details open_in_new

Journal of Cybersecurity and Information Management | 2026

Machine Learning-Driven Cyber Threat Prediction and Prevention: A Multi-Dataset Design and Comparative Evaluation

As technology advances, the frequency and variety of intrusions and other security threats within network environments continue to grow. Intrusion detection systems (IDS) play a vital role in securing networks against unauthorized access and attacks on computer systems; however, traditional IDSs are very limited in their ability to recognize new, complex malicious threats because they rely on signature-based detection. Approaches based on machine learning have shown a promising alternative in identifying unknown malicious attacks. This study proposes a computationally efficient, generalizable machine-learning framework for robust cyber-threat prediction. Three benchmark datasets (HIKARI-2021, CIC-IDS2017, and KDDCup99) were used for full-pipeline evaluations, including preprocessing, feature selection, class-imbalance handling, hyperparameter optimization, and strict model validation. Eight classifiers were assessed, which included traditional classifiers and more modern ensemble methods. The results from this study showed that tree-based models, mainly both Random Forest and XGBoost achieved near-perfect performance across all datasets, reaching accuracy values up to 0.999 and F1-scores between 0.99 and 0.999. Additionally, the SHAP-based explainability analysis was applied to reveal features that drove predictions, enabling interpretability and transparency. Compared with prior studies, the proposed framework consistently delivers improved, more stable detection performance. The findings highlight that optimized ML models combined with balanced datasets and rigorous validation protocols can significantly enhance intrusion detection reliability. Furthermore, this approach provides a practical and scalable solution for strengthening cybersecurity defenses against evolving and emerging cyber threats.

groups

Krishneel Sundar mail -

Pritika Reddy mail -

Kaylash C. Chaudhary mail

link https://doi.org/10.54216/JCIM.180106

Volume & Issue

Vol. Volume 18 / Iss. Issue 1

Details open_in_new

International Journal of Wireless and Ad Hoc Communication | 2026

TA-FaultNet: A Temporal Attention Framework with Bidirectional LSTM for Multi-Class Fault Detection and Health Monitoring in Industrial Wireless Sensor Networks

Industrial wireless sensor networks are central to the continuous monitoring of critical plant equipment, yet reliable identification of multiple concurrent fault modes from heterogeneous multivariate sensor streams remains an unsolved operational challenge. Physical failure mechanisms—pump cavitation, valve blockage, gradual sensor drift—and wireless channel disturbances each imprint distinct but overlapping temporal signatures that render classical thresholdand rule-based detectors inadequate for automated maintenance dispatch. This paper  presents TA-FaultNet, a neural architecture designed specifically for the multi-class fault identification problem in industrial sensor deployments. The network couples a two-stage stacked bidirectional recurrent encoder with a parallel multi-head self-attention module and a compact temporal convolutional block, enabling simultaneous capture of long-range process dynamics and fine-grained fault-onset localisation from raw sensor windows. TA-FaultNet is evaluated on the publicly available Skoltech Anomaly Benchmark under five operational classes and assessed through a comprehensive battery of experiments including baseline comparisons, systematic component ablation, cross-experiment generalisation, andprogressive noise-injection testing. The proposed architecture decisively outperforms eight competing methods spanning classical anomaly detectors, standalone recurrent and convolutional networks, and the Transformer, while remaining lightweight enough for edge gateway deployment. Attention weight visualisations expose fault-specific temporal activation patterns, providing maintenance engineers with interpretable diagnostic evidence beyond bare classification labels.

groups

Massila Kamalrudin mail -

Mustafa Musa mail

link https://doi.org/10.54216/IJWAC.100105

Volume & Issue

Vol. Volume 10 / Iss. Issue 1

Details open_in_new

International Journal of Wireless and Ad Hoc Communication | 2026

Event-Selective Fog Microbatching for Wireless Sensor IoT Devices: A Data-Driven Study Using Edge-IIoTset Features

Wireless sensor IoT devices increasingly operate under strict energy, latency, and security constraints while generating high-frequency telemetry that cannot be forwarded continuously to remote clouds. This paper presents an event-selective fog microbatching model for wireless sensor streams in which local novelty scoring, fog-side buffering, risk-preserving retention, and energy-aware scheduling are jointly optimized. Unlike conventional anomaly-detection pipelines, the proposed method treats communication reduction as a primary design objective and binds it mathematically to attack-evidence preservation. A reduced feature-level experimental file following the public Edge-IIoTset label structure and selected network/sensor attributes is used to evaluate traffic selectivity, uplink reduction, fog latency, energy saving, and detection performance. The model assigns each observation window a novelty score, suppresses redundant low-information traffic, and groups retained events into load-aware microbatches at the nearest fog node. The proposed model is extended with stochastic retention bounds, microbatch-delay stability, radio-energy equations, and risk-constrained threshold calibration. Experimental results show that the design reduces uplink load and radio-energy consumption while preserving strong attack discrimination across distributed wireless sensor traffic. The findings support a broader use of fog computing as a selective communication-control layer for dense, security-sensitive wireless sensor IoT deployments.

groups

Raden Aur Aachman Azakiyullah mail -

Aiswan Aumanti mail

link https://doi.org/10.54216/IJWAC.100201

Volume & Issue

Vol. Volume 10 / Iss. Issue 2

Details open_in_new

Journal of Cognitive Human-Computer Interaction | 2026

Guardian Light: An Edge-Resilient Fail-Safe Mechanism for IoT Smart Lighting Against DDoS and Network Partitions

New cybersecurity and operational resilience issues have been brought about by the growing use of cloud managedsmart street lighting in metropolitan settings, especially in the event of network partitioning and Distributed Denial of Service (DDoS) assaults. Current systems still rely mostly on centralized cloud control, which creates a single point of failure that might compromise public safety and interfere with vital lighting functions. In the context of the author’s Streetlight-as-a-Service (SLaaS) framework, where streetlights operate as intelligent, service-capable infrastructure nodes rather than discrete lighting devices, this paper proposes Guardian Light, an edge-resilient fail-safe mechanism for intelligent street lighting. The suggested design uses AWS IoT Core, AWS IoT Device Defender, and AWS IoT Greengrass to combine device-side autonomous governance with cloud-side anomaly detection. With the help of an internal real-time clock, state-aware failover logic, persistent offline scheduling, and local threshold monitoring, Guardian Light makes it possible for lighting nodes to continue operating safely and consistently even in the event that malicious traffic is discovered or cloud connectivity is compromised. The study emphasizes how current smart lighting research goes beyond energy saving and scheduling to cyber-resilient operational continuity through the integration of edge intelligence and service-oriented streetlight design. By doing this, the study offers a workable and theoretically sound solution to improve the autonomy, security, and dependability of next-generation SLaaS-enabled smart city systems.

groups

Lokman Fadzıl mail -

Tımothy Hong mail

link https://doi.org/10.54216/JCHCI.110201

Volume & Issue

Vol. Volume 11 / Iss. Issue 2

Details open_in_new

Journal of Cognitive Human-Computer Interaction | 2026

IHLawRecommender: Deep Semantic Modelling for IPC Case Recommendation with Legal Domain Constraints

Efficient retrieval of relevant legal cases is critical for judicial decision-making, particularly for high-severity crimes where timely reference to precedents can influence outcomes. Our work presents IHLawRecommender, i.e., Intelligent Hybrid Law Recommender, a hybrid framework for recommending Indian Penal Code (IPC) cases based on textual descriptions provided by users. The system operates through a multi-stage workflow: first, case descriptions are normalized to remove inconsistencies and embedded into semantic vectors using a Bi-directional Long Short-Term Memory (BiLSTM) network. These embeddings are compared with the user query to measure semantic similarity. In parallel, an IPC-specific keyword map evaluates the relevance of each case, while legal aware filters distinguish between sexual and non-sexual violent crimes to ensure contextually appropriate recommendations. The outputs from these stages are integrated using a weighted payoff function that considers semantic similarity, keyword relevance, and crime severity to produce a ranked list of top-k cases. The system also provides interpretable visualizations, including heatmaps that illustrate correlations between similarity, keyword score, severity, and payoff. Evaluation on a curated IPC dataset demonstrates that IHLawRecommender consistently prioritizes legally critical cases, reduces irrelevant matches, and offers a practical, workflow-driven tool for legal professionals to efficiently navigate case law while maintaining adherence to judicial priorities.

groups

Gautham Praveen Ramalingam mail -

Dharini Ramalingam mail -

A. Farhan mail

link https://doi.org/10.54216/JCHCI.110202

Volume & Issue

Vol. Volume 11 / Iss. Issue 2

Details open_in_new

Journal of Cognitive Human-Computer Interaction | 2026

Computer Resource Usability Modelling from Virtual-Machine Workload Traces: A Cognitive HCI Perspective

Computer usability is often discussed through screen layout, navigation, and task flow, although the experience of using a computer also depends on whether processor, memory, storage, and network resources remain available when the user needs them. This paper develops a Computer Resource Usability Index (CRUI) for interpreting virtual-machine resource traces as indicators of user-facing usability risk. The proposed index converts CPU, memory, disk, and network measurements into a bounded resource-friction score and then maps this score into four actionable usability states: comfortable, watch, constrained, and strained. The analysis uses a processed extract following the public GWA-T-12 Bitbrains trace structure, which records VM-level resource metrics for enterprise applications. The results show that resource usability is not explained by CPU usage alone; imbalance across resource channels, I/O pressure, and variability also contribute to predicted friction. The findings provide a practical bridge between infrastructure monitoring and cognitive HCI by translating low-level resource traces into interface-relevant decisions such as when to defer background tasks, warn the user, or allocate additional headroom.

groups

Fadi Farha mail -

Tony Salloom mail

link https://doi.org/10.54216/JCHCI.110203

Volume & Issue

Vol. Volume 11 / Iss. Issue 2

Details open_in_new

Journal of Cognitive Human-Computer Interaction | 2026

Adaptive Interface Personalization through Real-Time Cognitive Load Detection

High-stakes computer work often requires users to interpret dense visual information while responding to timesensitive events. Static interfaces can become counterproductive in such conditions because the amount of information presented to the user does not change when mental demand rises. This paper presents an adaptive interface personalization approach that detects cognitive load from pupillometry, heart-rate variability, gaze behaviour, and interaction traces, then selects a transparent interface response. The proposed approach does not simply reduce screen content; it chooses between full, highlighted, simplified, and critical-only modes while preserving user control and explanation cues. A feature-level experimental analysis was conducted using a multimodal workload table structured around public cognitive-load datasets and high-stakes monitoring tasks. The results show that pupil expansion, lower HRV, response delay, gaze dispersion, and screen density jointly indicate rising cognitive load. The adaptation policy reduced predicted interaction errors and shortened response latency in high-load windows while maintaining explanation support for user trust. The findings suggest that cognitive-load detection should be treated as a personalization service rather than a hidden automation layer.

groups

Wadhah Abdullah mail -

Aygul Z. Ibatova mail

link https://doi.org/10.54216/JCHCI.110204

Volume & Issue

Vol. Volume 11 / Iss. Issue 2

Details open_in_new

Journal of Cognitive Human-Computer Interaction | 2026

Explaining AI Decisions to Mitigate Cognitive Biases in Human-AI Collaboration

Human-AI collaboration can improve decision quality only when users know when to rely on an AI recommendation and when to resist it. Explanations are often proposed as a remedy, but explanation content can also intensify automation bias or reinforce a user’s initial belief. This paper presents a cognitive explanation selection model for mitigating over-reliance and under-reliance in AI-assisted decision tasks. The study compares no explanation, feature-based, contrastive, example-driven, and hybrid explanations across simulated novice, intermediate, and expert decision makers using a public medical decision dataset as the task substrate. The analysis focuses on reliance behaviour rather than on model accuracy alone. The proposed model estimates when the user is likely to accept a wrong recommendation, reject a correct recommendation, or accept advice simply because it confirms an initial judgment. The results indicate that contrastive and hybrid explanations are more effective for reducing automation bias, while example-driven explanations preserve trust for lower-expertise users. The paper concludes with a transparent interface loop for high-stakes environments in which explanation style is selected according to user expertise, AI confidence, and human-AI agreement.

groups

Aiswan Aumanti mail -

Citra Dewi mail

link https://doi.org/10.54216/JCHCI.110205

Volume & Issue

Vol. Volume 11 / Iss. Issue 2

Details open_in_new