A novel approach for missing data recovery and fault nodes detection in wireless sensor networks

Abstract

Ensuring data integrity in wireless sensor networks (WSNs) is crucial for accurate monitoring, yet missing data due to sensor faults present a significant challenge. This research introduces an innovative approach that integrates advanced data recovery techniques with leading-edge methods to address this issue. The system begins by identifying and isolating fault nodes using a specialized algorithm that analyzes network behavior. By applying fuzzy density-based spatial clustering of applications with noise (FDBSCAN), potential fault nodes are precisely located based on deviations from expected patterns. Subsequently, an intelligent missing data recovery mechanism powered by bidirectional long short-term memory (Bi-LSTM) networks takes action. The Bi-LSTM model is trained on existing sensor data to capture intricate patterns and dependencies, enabling accurate prediction and reconstruction of missing values caused by identified faults. The synergy between Bi-LSTM for missing data recovery and FDBSCAN for fault node detection comprehensively addresses the missing data problem in WSNs. In missing data recovery, it demonstrates low mean absolute deviation (MAD) ranging from 0.021 to 0.13 and mean squared deviation (MSD) ranging from 0.0025 to 0.05 across various missing data ratios. Data reliability remains consistently high at 96% to 98%, even with up to 80% missing data. For fault node detection, the approach achieves precision of 95.7%, recall of 96.3%, F1-score of 96.1%, and accuracy of 97.4%, outperforming existing techniques. The computational cost during training is noted at 5.79 h, presenting a limitation compared to other methods. This research highlights the importance of integrating fault node detection into missing data recovery mechanisms, presenting an innovative solution for the advancement of WSNs.