|
A comparison of independent variable selection in a logistic regression model using Bayesian variable selection and stepwise regression |
|---|---|
| รหัสดีโอไอ | |
| Creator | Kannat Na Bangchang |
| Title | A comparison of independent variable selection in a logistic regression model using Bayesian variable selection and stepwise regression |
| Contributor | Phattharaphon Srisod, Methapohn Sarasri, Phuripat Suksee |
| Publisher | Mahasarakham University |
| Publication Year | 2569 |
| Journal Title | Journal of Science and Technology Mahasarakham University |
| Journal Vol. | 45 |
| Journal No. | 3 |
| Page no. | 348-356 |
| Keyword | Logistic regression, multicollinearity, Bayesian variable selection, stepwise regression, Gibbs sampling |
| URL Website | https://li01.tci-thaijo.org/index.php/scimsujournal |
| Website title | Journal of Science and Technology Mahasarakham University |
| ISSN | 1686-9664 (Print), 2586-9795(Online) |
| Abstract | Selecting appropriate independent variables yields a highly efficient model, particularly in regression analysis. This study aims to examine the selection of independent variables in a logistic regression model using Bayesian variable selection via Gibbs sampling and stepwise regression. It compares these two selection methods under conditions of very low and very high multicollinearity among independent variables. The study is conducted through data simulation and applied to online writing behavior data for diagnosing Alzheimer's disease. The sample sizes for the simulation were set to 25 and 100, with 100 replications for each case. The performances of Gibbs sampling and stepwise regression were compared based on evaluation criteria, including the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), number of correctly selected independent variables, average estimated coefficients of correctly selected variables, and accuracy percentage. The results indicate that for both small (n = 25) and large (n = 100) sample sizes, when the multicollinearity among independent variables is very low, Gibbs sampling and stepwise regression show no significant difference in their efficiency in selecting independent variables. However, in cases where multicollinearity is very high, Gibbs sampling demonstrates superior performance in selecting independent variables compared to stepwise regression. The findings of this research can be applied to select independent variables in real-world data exhibiting multicollinearity. |
