Incremental unsupervised learning using clustering
The concept behind clustering in a data stream remains the same as in batch or offline modes; that is, finding interesting clusters or patterns which group together in the data while keeping the limits on finite memory and time required to process as constraints. Doing single-pass modifications to existing algorithms or keeping a small memory buffer to do mini-batch versions of existing algorithms, constitute the basic changes done in all the algorithms to make them suitable for stream or real-time unsupervised learning.
Modeling techniques
The clustering modeling techniques for online learning are divided into partition-based, hierarchical-based, density-based, and grid-based, similar to the case of batch-based clustering.
Partition based
The concept of partition-based algorithms is similar to batch-based clustering where k clusters are formed to optimize certain objective functions such as minimizing the inter-cluster distance, maximizing...
