Detection of SPAM attacks in the remote triggered WSN experiments

Spam attack is the deliberate delivery of unsolicited or unwanted messages across the computer networks with the intention to deplete the resources that results in Denial of Service (DoS) to the end user. This is more important to consider in Wireless sensor networks test beds where the nodes already have only little computing resources (4kb RAM), and low network bandwidth for their applications. The Remote Triggered WSN test bed (http://vlab.amrita.edu/?sub=78) that we have deployed in our university consists of more than 80 nodes connected with various sensors, digital multimeters etc., allows any student in the internet to upload their programs, execute them and view their experiment results with real time video streaming to learn the WSN concepts intuitively. Hence, there is a need to detect such type of spam attacks in the test bed, in case, a user uploads the malicious programs that affects the functioning of nodes in other experiments. We have tried two packet inspection techniques, Gaussian Naive Bayes (GNB) and k-Nearest Neighbour (K-NN) for learning the pattern and identifying whether the new incoming message is Spam or Non-spam. It is observed that the GNB method could catch spam messages at 94-96% Accuracy, with only 5-10% false positive rate (FPR). It is also found that the performance of k-NN gradually decreases as k-value increases. The complexity and execution speed becomes worse at larger k-values where as they are invariant in case of GNB. Hence it shows GNB is more appropriate than k-NN for inspecting the messages.