Abstract
We introduce a statistic testing for neglected nonlinearity using extreme learning machines and call it ELMNN test. The ELMNN test is very convenient and can be widely applied because it is obtained as a by-product of estimating linear models. For the proposed test statistic, we provide a set of regularity conditions under which it asymptotically follows a chi-squared distribution under the null. We conduct Monte Carlo experiments and examine how it behaves when the sample size is finite. Our experiment shows that the test exhibits the properties desired by our theory.
| Original language | English |
|---|---|
| Pages (from-to) | 117-129 |
| Number of pages | 13 |
| Journal | International Journal of Uncertainty, Fuzziness and Knowlege-Based Systems |
| Volume | 21 |
| Issue number | SUPPL.2 |
| DOIs | |
| Publication status | Published - 2013 Dec |
Bibliographical note
Funding Information:The authors are most grateful to the editors and three anonymous referees. We are also benefited from discussions with Halbert White and the participants of ELM2012 held at Singapore. Jin Seo Cho acknowledges research support from the Korea Sanhak Foundation.
All Science Journal Classification (ASJC) codes
- Software
- Control and Systems Engineering
- Information Systems
- Artificial Intelligence