International Journal of Emerging IoT Technologies in Smart Electronics and Communication

Reinforcement Learning for Adaptive Congestion Control in IoT Networks: A Deep Q-Network Approach

Maloani Saidi Georges — 2026-08-24

This study presents a Reinforcement Learning (RL)-based approach for congestion control in Internet of Things (IoT) networks, addressing the limitations of traditional methods such as Transmission Control Protocol (TCP). The proposed system leverages a Deep Q-Network (DQN) model to dynamically adapt transmission strategies based on real-time network conditions. A simulation framework was developed using Python, integrating PyTorch for the RL model, NumPy for numerical computations, Matplotlib for visualization, and PyQt5 for an interactive dashboard. The performance of the RL-based approach was evaluated against TCP using key metrics, including throughput, latency, packet loss, and energy consumption. The results demonstrate that the RL model significantly enhances network performance, achieving higher throughput while reducing latency and packet loss. Additionally, the approach shows improved energy efficiency, making it suitable for resource-constrained IoT environments. The findings confirm that RL-based congestion control provides a more adaptive, efficient, and intelligent solution compared to conventional techniques. This work contributes to the development of next-generation IoT systems by integrating artificial intelligence into network management, offering promising applications in smart cities, healthcare, and intelligent transportation systems.

A Unified Explainable Transfer Learning Framework for Robust Cross-Domain Malware Analysis and IoT Network Intrusion Detection

Belay Sitotaw Goshu — 2026-09-23

The proliferation of sophisticated cyber threats targeting both traditional endpoints and Internet of Things (IoT) devices requires unified detection frameworks that operate across heterogeneous domains while maintaining interpretability. This paper presents XPLAIN-Transfer, a novel framework integrating deep transfer learning with SHAP-based explainable AI (XAI) for robust cross-domain malware analysis and IoT network intrusion detection. The methodology employs a hybrid CNN-LSTM architecture pre-trained on the Malware Memory Dump dataset and fine-tuned for IoT (NF-TON-IoT) and network (UNSW-NB15) intrusion detection through a three-phase transfer learning strategy. Extensive benchmarking shows that XPLAIN-Transfer consistently outperformed conventional training approaches, achieving accuracies of 99.9%, 96.2%, and 96.0% for malware, IoT intrusion, and network intrusion detection, respectively, with performance gains ranging from 5.7% to 7.7%. Transfer learning reduces training time by 40% - 51%, while SHAP integration provides transparent, actionable explanations for security analysts. Robustness validation confirms resilience against adversarial attacks and concept drift. The findings establish that unified explainable frameworks can effectively bridge memory-based and network-based threat detection domains, offering practical deployment advantages to heterogeneous environments. XPLAIN-Transfer represents a significant advancement toward interpretable, efficient, and cross-domain cybersecurity solutions.

Design and Development of an Intelligent IoT-Based Air Quality Monitoring and Alert System, Challenges, Ethics, and Future Perspectives

Suraj R. Nalawade — 2026-09-28

Air pollution poses a serious threat to human health and the environment, making continuous monitoring and timely warning systems essential for protecting public health. This article presents the design and development of an integrated Air Quality Monitoring and Alert System (AQMAS) that monitors particulate matter (PM2.5 and PM10), Carbon Dioxide (CO₂), Nitrogen Dioxide (NO₂), and Ozone (O₃) in real time using low-cost, easy-to-deploy sensor nodes. The system architecture comprises sensor nodes, an Arduino Uno microcontroller unit (with ESP32/Raspberry Pi noted as potential platforms for future wireless-communication upgrades), a cloud platform for data storage and visualization, and a multi-channel alert mechanism. Sensor readings are filtered using a moving-average technique, converted into standardized Air Quality Index (AQI) values based on EPA/WHO guidelines, and transmitted to a cloud dashboard for real-time monitoring. When pollutant concentrations exceed predefined safety thresholds, the system automatically dispatches alerts through SMS, mobile app notifications, and on-site LED/buzzer indicators, enabling both individuals and health authorities to respond promptly. Field testing showed PM2.5 concentrations ranging from 12 to 154 µg/m³ (mean 55 µg/m³), PM10 from 20 to 210 µg/m³, and CO₂ from 420 to 1100 ppm, with pollution peaks consistently observed during morning and evening traffic hours. The alert system successfully delivered real-time notifications within an average of 10 min of threshold breach. The results confirm that low-cost, IoT-enabled sensor networks can provide reliable, community-scale air quality monitoring and demonstrate strong potential for deployment in homes, schools, hospitals, and smart-city initiatives.