Certain investigations on improving outlier detection accuracy in wireless sensor networks

Abstract

Nowadays, wireless sensor networks (WSNs) are gaining popularity in a newlinevariety of civilian and military applications. In WSN, data fusion or data newlineaggregation is carried out to gather data in a cluster head (CH) from other nodes newlinein that cluster. These aggregated data will be forwarded to the base station for newlineanalysis. In a lively environment, WSN data that is measured and gathered may newlinebe sometimes inaccurate. The issues about data reliability in WSNs imply that the newlinesensor data can be inaccurate, which further affects the quality of the raw data and newlinethe aggregated results that are forwarded to the base station for analysis. newlineAdditionally, transmitting inaccurate data to the base station consumes needless newlinebattery power, reducing the lifespan of the network. Identification of abnormal newlineor inaccurate data that vary intensely from remaining data readings is considered newlinesuspicious that needs to be focused on and researched. This issue is addressed newlineas Outlier detection (OD) which performs the classification of normal data from newlineabnormal data. Implementing OD in WSNs aids in removing inaccurate data newlinetransmission from CH to the base station which is further considered in this newlineresearch work. newlineThe gathered sensor data may be imbalanced where the abnormal newlineinstances are available in minimum amounts. When dealing with imbalanced newlinedata, the OD system can suffer from yielding better detection accuracy and more newlinefalse positives than false negatives. This challenging task motivated researchers newlineto build an OD model to improve detection accuracy with less computation newlinecomplexity and a reduced number of false alarms. This attracted researchers to newlinedevelop an efficient OD model that should be able to classify abnormal newlineinstances with high detection accuracy and reduce false alarms. In recent years, newlineunsupervised outlier detection (UOD) using deep learning has proved to newlineimprove classification accuracy when dealing with imbalanced data produced newlineby various real-time applications newline