Time series forecasting is an important research topic applied to various areas of human knowledge, such as economics, medicine, meteorology, and engineering. The forecasting results can assist the decision-making process, providing useful projections to specialists. In this paper, a hybrid approach named Parallel Recursive Singular Spectrum Analysis and Evolving Neuro-Fuzzy Network (PRSSA+ENFN) for forecasting univariate and multivariate experimental time series, with learning from data streams composed of time series samples, is presented. The adopted methodology considers an evolving neuro-fuzzy basis, i.e., the PRSSA+ENFN adapts, extends, and evolves the fuzzy rule structure from new incoming data of the time series. The time series is subdivided into unobservable components, which are patterns contained in the data, and the neuro-fuzzy network forecasts these component samples so to reconstruct the original time series considering future values. An alternative to determine the specific parameters of PRSSA+ENFN is the application of a genetic algorithm. Four multi-steps ahead forecasting experiments are outlined with univariate and multivariate time series. Finally, some competitive results in relation to other methods in literature are discussed.