Development of the industrial Internet concept dictates the need for identification and improvement of approaches, models, and methods for analyzing the state of the Internet of Things. Implementation of modern industrial, social, and household systems is impossible without the use of artificial intelligence methods in the machine-to-machine communication of individual elements, automatic data collection, analysis, and storage. The paper presents an approach to identifying the state of devices based on the application of classification technology, which implements compositions of independently trained algorithms processing time series, reflecting the functioning of elements during the implementation of processes. The application of the proposed solution allows parallel processing of information received from the device, which enables scaling. The developed approach was tested on time series sequences, obtained experimentally in different operating conditions, and processed by a sequence of classifiers. The paper presents the results of the probability estimate of erroneously classified states. The main advantages of the proposed solution are relatively small requirements to computational resources, simplicity of implementation, and the ability to scale by adding new classification algorithms.