An efficient adaptation of the basic windowing technique shown in Figure 1 to noisy domains is a non-trivial endeavor. In particular, it cannot be expected that the use of a noise-tolerant learning algorithm inside the windowing loop will lead to performance gains in noisy domains. In our opinion, the main problem with windowing in a noisy domain lies in the fact that a good theory will misclassify most of the noisy examples, and consequently incorporate them into the learning window for the next iteration. On the other hand, the window will typically only contain a subset of the original learning examples. Hence, after a few iterations, the proportion of noisy examples in the learning window can be much higher than the noise level in the entire data set. Naturally, this makes the task for the learning module considerably more difficult.
Assume, for example, that your favorite noise-tolerant learner has learned a correct theory from a randomly selected starting window of size 1000 in a 11,000 examples domain. Further assume that 10% of the examples are labeled incorrectly. Therefore, the correct theory will misclassify 1000 of the remaining 10,000 examples because they are noisy. These examples will consequently be added to the window, thus doubling its size. Assuming that the original window also contained about 10% noise, more than half of the examples in the new window are now erroneous, so that the classification of the examples in the window is in fact random. It can be assumed that many more examples have to be added to the window in order to recover the structure that is inherent in the data. This conjecture is consistent with the experimental results of  and , which showed that windowing is highly sensitive to noise.