Knowledge discovery process is to provide effective rules from the existing data. The process of finding knowledge is a complex task when the data source is in the form of data streams and in fact the data source of class imbalance nature. To find the opt solutions for those problems research proposals are formulated by many researchers. The some of the unsolved problems in the literature for the above said problem is for very large data sources of data streams with class imbalance nature. In this paper, a novel proposal for class imbalance large data streams is presented with novel techniques of oversampling and a unique multi modal filtering technique and oversampling strategy as Multimodal Increment Over Sampling for Data Streams (MIOSDS) for efficient implementation of proposal for a better solution. The experimental simulations are conducted on three large datasets with different domains. The results generated are good compared with benchmark evaluation metrics.