In this research, we introduce a unique and efficient sliding window model-based discriminative item set mining technique over data streams. Item sets classified as discriminative are frequently found, and they occur there far more frequently than they do in the other streams. More difficulties arise when mining discriminative item sets than when mining frequent ones, particularly in the sliding window model where the algorithms must handle item set in multiple data stream during window frame sliding in order to process transaction coming and going the window. We suggest utilizing two new in-memory data structures in a single scan technique to mine discrimination item set in a mixed window setting. Controlling the creation of several unpromising item sets is done by offline processing. Between two offline sliding, online processing is performed to obtain more accurate and current online replies. The online mining operation is carried out in between two periodic offline sliding windows, where the discriminative item sets that have been found are accurately updated. The thorough empirical research demonstrates that the suggested algorithm accurately and efficiently provides time and space complexities. Large, swift, and complicated data streams can be handled by the algorithm.