Experimental comparison of the diagnostic capabilities of classification and clustering algorithms for the QoS management in an autonomic IoT platform

The Internet of Things (IoT) platforms must allow the communication between the Applications and Devices according to their non-functional requirements. One of the main non-functional requirements is the Quality of Service (QoS). In a previous work has been defined an autonomic IoT platform for the QoS Management, based on the concept of autonomic cycle of data analysis tasks. In this platform have been defined two autonomic cycles, one based on a classification task that determines the current operational state to define the set of tasks to execute in the communication system to guarantee a given QoS. The other one is based on a clustering task that discovers the current operational state and, based on it, determines the set of tasks to be executed in the communication system. This paper analyzes the diagnostic capabilities of the system based on both approaches, using different metrics. For that, a real scenario has been considered, with simulations that have generated data to test both tasks. Each technique has different aspects to be considered for a correct QoS management in the context of IoT platforms. The classification technique can determine very well the learned operational states, but the clustering approach can carry out a more detailed description of the operational states. Additionally, due to the classification and clustering technique used, called learning algorithm for multivariate data analysis, the paper analyzes the operational state profile determined by them, which is very useful in a diagnostic process.