Abstract Feature selection presents many challenges and difficulties during online learning. In this study, we focus on fuzzy feature selection for fuzzy data stream. We present a novel incremental feature weighting method with two main phases comprising offline fuzzy feature selection and online fuzzy feature selection. A sliding window is used to divide the fuzzy data set. Each fuzzy input feature is assigned a weight from [ 0 , 1 ] according to the mutual information shared between the input features and the output feature. These weights are employed to access the candidate fuzzy feature subsets in the current window and based on these subsets, the offline fuzzy features selection algorithm is applied to obtain the fuzzy feature subsets by combining the backward feature selection method with the fuzzy feature selection index in the first sliding window. The online feature selection algorithm is performed in each of the new sliding windows. The feature subset in the current window is updated by combining the fuzzy feature selection results from the previous sliding window with the current candidate fuzzy feature set according to the importance level of the fuzzy input feature. Finally, the evolving relationships of the fuzzy input features are found using the fuzzy feature weight between the sliding windows. Simulation results showed that the proposed algorithm obtains significantly improved adaptability and prediction accuracy compared with existing algorithms.