Convergence detection in classification task of knowledge discovery process

The adaptive incremental approach to classification task of data mining has a built-in feature to detect convergence of a classification algorithm. The feature is given in form of three equations, which must be all fulfilled. The equations are parametric and can be modified based on miner's personal experiences with the dataset at hand or similar datasets. The advantages of using the approach are potentially lower data preparation costs, lower algorithm execution times, good insight into the algorithm's behavior based on small subset of data, and possibility to predict algorithm's final error rate or based on the desired final error rate, to predict sample size to obtain it. In the future, the authors plan to validate their model on additional datasets and with several other data mining algorithms that build models and produce error rates. Additionally, they plan to incorporate the (run) time component into their framework.