ABSTRACT
Robust imputation approaches are required due to the ubiquitous problem of missing data in WSNs, which might be caused by failures, maintenance, or communication outages. When faced with repeated missing data points or when depending on other simultaneously monitored characteristics, existing techniques frequently break down. Deep learning techniques are promising, but they require a lot of training data, which is hard to come by in wireless sensor net- works. We suggest an updated SSIM to address these issues. SSIM combines past and future data for a given period by utilizing cutting-edge sequence- to sequence deep learning architecture, particularly LSTM networks. To further enable effective SSIM training even with minimal data, we offer a VSFW technique to obtain adequate training examples. An assessment using actual time series data from a network of water quality monitors shows how much better SSIM is than existing methods. Our SSIM yields significant improvement of performance in terms of error parameters MAE, RSME, RSMLE, and DTW for recovering missing data sequences of varying lengths. These results highlight SSIM's bright future for improving data quality management in WSN.
