Modelling the relationship between problem characteristics and data mining algorithm performance using neural networks, in: C. Dagli, et al

As more data mining algorithms become available, the answer to one question becomes increasingly important: which techniques are best suited to which data mining problems? This study attempts to address this question by examining the performance of six leading data mining algorithms across a collection of 57 well-known classification problems from the machine learning repository. For each data set, a number of statistics are collected in order to measure the size and complexity of the problem. Using a neural network, the performance results of the six algorithms are then combined with the problem characteristics to build a predictive model of the relative performance of each algorithm on a given problem