MALADY: A Machine Learning-Based Autonomous Decision-Making System for Sensor Networks

As the capabilities of sensor networks evolve, we need to address the challenges that will help in shifting the perception of sensor networks as being merely a data-gathering network to that of a network that is capable of learning and making decisions autonomously. This shift in intelligence from the edge to the nodes in the network is particularly relevant in unattended sensor deployments where there is no continuous access to a remote base station. In this paper, we propose an architecture, called MALADY, which uses a machine learning approach to enable a network of embedded sensor nodes to use the data that they have gathered to learn and make decisions in real-time within the network and thereby, become autonomous. MALADY supports supervised as well as unsupervised learning algorithms. Our implementation of the algorithms introduces some practical optimizations, in order to make them viable for nodes with limited resources. Our experimental results base don an implementation on the MicaZ mote/TinyOS platform show that the supervised learning technique based on linear discriminant analysis has a higher learning complexity, but allows a sensor node to learn about the data correlations robustly and make decisions accurately, after learning from only a few samples. In comparison, the unsupervised learning technique based on clustering has a low overhead, but requires more learning samples to achieve a high detection accuracy.