Full Program »
Maximizing Sensors Trust Through Support Vector Machine
Trust has become a major concern in wireless sensor networks (WSN), since many WSNs are deployed in many data sensitive applications, especially in healthcare and surveillance. Factors, such as the sensitivity to internal and external noises, constraints on sensing devices, varying deployment platforms and network topologies, generate uncertainty in data accuracy and consistency. With the advent of smart sensing solutions, the data accuracy is increased to a greater extent. However, explicit discrimination of uncertainty from the sensor data is still remain challenging, as it is difficult to quantify the individual impact. The environmental uncertainty is one of the difficult parameter to quantify and to predict, and it greatly influences the received signal strength (RSS) in outdoor sensor networks, leveraging frequent data inconsistency ended up with sensor distrust. We propose a framework to maximize sensors’ trust by classifying the level of impact on the existence of a few environmental uncertainties, such as temperature, humidity and wind speed. We have applied multiclass support vector machine (SVM) classifier to analyze RSS of sensor under the individual and combined presence of defined environmental uncertainties and to classify the dataset into four groups; moreover, the penalty corresponding to the level of uncertainty is added to boost the sensor trust. We have selected Quadratic SVM to train the dataset, as the data varied non-linearly. The experiment shows 97% accuracy during training and 96.2% accuracy during testing with 3.8% misclassifications. With these predicted level of uncertainties and corresponding boosting in RSS, the framework is found to move 42% of sensors from uncertain to trusted category.