This paper presents an analysis of parameter tuning for induction algorithm oriented feature elimination (IAOFE), an approach that takes into consideration not only the data and the target concept, but also the induction algorithm that will learn the target concept from the data. Because of its very nature, IAOFE is controlled by abounding parameters. It would be of great utility if one knows what parameter settings can inspire the ideal performance out of IAOFE. Unfortunately, little work has

more »

... een done to address this issue. This paper aims at filling this blank area on a comprehensive scale. Parameters have been defined and explained. Comparative studies have been conducted for various parameter settings. Effective configurations of parameter settings have been identified. Empirical evidence from a large number of datasets demonstrates that IAOFE with the suggested parameter configurations can achieve higher predictive accuracy than existing popular feature selection approaches with statistically significant frequencies.