A Quick Outlier Detection in Wireless Body Area Networks

AUDIT: AnomaloUs data Detection and Isolation approach for mobile healThcare systems

“…Moreover, other approaches (Salem, Liu, et al, ; Shravan et al ) consider only spatial correlations between physiological attributes without taking into account the temporal correlation, hence having a low detection accuracy and high computational complexity in terms of time and storage requirements that are not available in a constrained device (e.g., smartphone). 3.The majority of approaches, except Aderibigbe and Chi () and Salem, Liu, et al (), require more effort in the data collection phase because they need a trusted learning (training or preprocessing) data set to precisely construct classification, regression, or clustering models. Learned models are then used to detect anomalies.…”

“…Aderibigbe and Chi () addressed the issue of anomaly detection by proposing a centralized approach in wireless body area networks. The proposed approach is based on the following metrics: median absolute deviation filter for temporal anomaly detection and majority voting for decision purposes.…”

“…On the other hand, multivariate anomaly detection approaches are classified according to their corresponding techniques to find classes: classification-based (Salem, Guerassimov, et al, 2013;Salem, Guerassimov, et al, 2014;Shravan et al, 2017;Zhang et al, 2015), clustering-based (Naseem et al, 2013;Elmougy et al, 2017), nearest neighbour-based Yang et al, 2015), statistical-based (Saneja & Rani, 2017), and spectral decomposition-based (Deneshkumar, Senthamaraikannan, & Manikandan, 2014;. Aderibigbe and Chi (2017) addressed the issue of anomaly detection by proposing a centralized approach in wireless body area networks. The proposed approach is based on the following metrics: median absolute deviation filter for temporal anomaly detection and majority voting for decision purposes.…”

“…3. The majority of approaches, except Aderibigbe and Chi (2017) and Salem, Liu, et al (2013), require more effort in the data collection phase because they need a trusted learning (training or preprocessing) data set to precisely construct classification, regression, or clustering models. Learned models are then used to detect anomalies.…”

“…Some solutions (Elmougy, Hossain, Tolba, Alhamid, & Muhammad, ; Shravan, Vaidehi, Sangeetha, Antony, & Vaikundam, ) detect data anomalies in the server side. However, the other solutions (Ahsanul, Mustafizur, & Mahfuzul, ; Aderibigbe & Chi, ; Naseem, Salem, Liu, & Mehaoua, ; Saneja & Rani, ; Salem, Liu, & Mehaoua, ; Salem, Guerassimov, Mehaoua, Marcus, & Furht, ; Salem, Liu, & Mehaoua, ; Salem, Liu, Mehaoua, & Boutaba, ; Salem, Guerassimov, Mehaoua, Marcus, & Furht, ; Yang, Liu, Gao, Qiu, & Meng, ; Zhang, Liu, & Li, ) are designed to be deployed in the personal processing unit to minimize the transmission of erroneous measurements to the server and hence increase the system lifetime. Furthermore, existing detection approaches are based either on univariate techniques (Aderibigbe & Chi, ; Ahsanul et al, ; Salem et al, ) or on multivariate techniques (Salem, Guerassimov, et al, ; Salem, Liu, et al, ; Salem, Liu, et al, ; Salem, Guerassimov, et al, ; Naseem et al, ; Saneja & Rani, ; Yang et al, ; Zhang et al, ).…”