Performing online feature subset selection (OFS) when data samples arrive at a high velocity in a time-dependent manner is a critical problem to solve. The situation becomes more difficult when the features also arrive in a stream. Several efforts were made by the researchers to perform OFS over feature streams. However, they are not scalable and cannot analyze feature streams coming at a high velocity. Further, optimal feature subsets must be identified by scalable approaches. It is noteworthy that evolutionary algorithms (EAs) which are inherently parallel and scalable are least employed for OFS in a streaming feature case. This motivated us to address the challenges and propose a generic EA-based wrapper for OFS to mine feature streams under the Apache Spark environment.