The Internet of Things (IoT) has emerged from the proliferation of mobile devices and objects connected, resulting in the acquisition of periodic event flows from different devices and sensors. However, such sensors and devices can be faulty or affected by failures, have poor calibration, and produce inaccurate data and uncertain event flows in IoT applications. A prominent technique for analyzing event flows is Complex Event Processing (CEP). Uncertainty in CEP is usually observed in primitive events (i.e., sensor readings) and rules that derive complex events (i.e., high-level situations). In this paper, we investigate the identification and treatment of uncertainty in CEP-based IoT applications. We propose the DST-CEP, an approach that uses the Dempster–Shafer Theory to treat uncertainties. By using this theory, our solution can combine unreliable sensor data in conflicting situations and detect correct results. DST-CEP has an architectural model for treating uncertainty in events and its propagation to processing rules. We describe a case study using the proposed approach in a multi-sensor fire outbreak detection system. We submit our solution to experiments with a real sensor dataset, and evaluate it using well-known performance metrics. The solution achieves promising results regarding Accuracy, Precision, Recall, F-measure, and ROC Curve, even when combining conflicting sensor readings. DST-CEP demonstrated to be suitable and flexible to deal with uncertainty.
Smoothed Deep Neural Networks for Marine Sensor Data Prediction
