PulseExploreJournal ClubDebatesTrendingResearchersJournals
Instagram
HomeExploreJournal ClubTrending
Synapse
⌘+K
Synapse
February 22, 2026Discrete Mathematics Algorithms and Applications0 citations

HADrTCN: Drift-Based Attention Enabled Deep Learning Model for Anomaly Detection in Internet of Things-enabled Smart Home Applications

View Full Paper
MMMuneeruddin MohammedMIMuhammad IrshadMKMohd. Nasair Uddin Khan

Key Points

  • The study aims to develop an improved model for anomaly detection in IoT networks to address limitations in conventional methods.
  • Developed the HADrTCN model incorporating hierarchical attention and drift-based tuning.
  • Implemented parallel layers for enhanced processing speed and scalability.
  • Utilized Generative Adversarial Network for handling dataset imbalances and improving accuracy.
  • Conducted data preprocessing to ensure feature normalization and consistency.
  • Achieved an anomaly detection accuracy of 98.13% on the CIC-IDS2017 dataset.
  • Demonstrated a 1.94% improvement over existing HEET and a 1.07% improvement over TSMAE.
  • Showed that the model effectively adapts to evolving data patterns, enhancing detection reliability.

Abstract

The adoption of the Internet of Things (IoT) has increased across diverse domains, where home appliances have recently been equipped with sensors for internet connectivity. However, the generation of large network traffic has led to the occurrence of anomalous activities. The conventional methods designed for anomaly detection lack full-spectrum monitoring and lead to false detection, resource wastage, and biased detections in high-dimensional data. Therefore, this research proposes the Hierarchical Attention-enabled Drift-based Tunable Convolutional Neural Network (HADrTCN) model to address the prevailing challenges as well as to provide accurate anomaly detection. Further, the drift-based tunable network automatically adapts to evolving data patterns via dynamically adjusting the parameters to detect the anomaly patterns in IoT networks. In addition, the HADrTCN model is distributed as two parallel functioning layers that promote faster processing as well as improve the scalability in handling larger datasets. Moreover, the Hierarchical Attention (HA) layer is incorporated in each distributed layer, allowing the proposed model to focus on the key discriminative features for detecting the anomaly patterns. Furthermore, the Generative Adversarial Network (GAN) is leveraged to deal with the dataset imbalances through refining the class distributions. Data preprocessing is performed to resolve the inconsistencies as well as to normalize the feature attributes. Experimental results show that the HADrTCN model outperforms the other existing techniques by achieving the high accuracy of 98.13%, indicating the substantial improvement of 1.94% compared to existing HEET, and 1.07% compared to TSMAE on CIC-IDS2017 dataset.

Ask AI
Helpful
Bookmark
Share
View Full Paper

Cite This Study

Mohammed et al. (2026) studied this question.

synapsesocial.com/papers/699a9d8e482488d673cd3735https://doi.org/10.1142/s1793830926500229
Ask AI
Helpful
Bookmark
Share
View Full Paper