This paper presents a time series classification method which utilizes ensemble learning. The multi-class classification is reduced to several binary classification tasks. Base classifiers incorporate a fuzzy cognitive map. Each of them learns independently to correctly associate time series with specific categories. Predicted memberships are later aggregated into the final class assignment. We compare four different decomposition methods implementing “one-vs-one” and “one-vs-all” approaches. We test these methods on real-world datasets and compare them with state-of-the-art solutions.