Cost-sensitive, Scalable and Adaptive Learning Using Ensemble Methods

Prior work in inductive learning focused on generic§algorithms that sought to reduce complexity. Thus,§simplifying assumptions were made. 1. All data§resides on a single processor, and resides§entirely in main memory;§Clearly in modern organizations today, most data§resides in a distributed§architecture with only small portions being resident§in main memory at each moment. §2. Each datum is considered equally important and uniform§costs are assumed. In real world contexts, different§exemplars frequently have§varying costs.§3. All features are freely acquired with no§computational or monetary§costs. This is unrealistic for many applications,§such as medical diagnosis. Usually, the test for each§feature consumes different costs and cannot be§ignored, i.e., accurate models that only take§advantage of the most expensive features are not§acceptable. §4. Model is computed on the basis of complete§knowledge. A learned hypothesis will§be applied to scenarios that are completely§represented in the training set. This§assumption is more often violated than satisfied.§There are usually new and unknown§patterns that traditional hypotheses will either§ignore or misclassify.