|
主讲人 |
廖军 |
简介 |
<p>Modal regression is a vital statistical modeling technique that focuses on the conditional mode. However, the models we used are usually misspecified, which may lead to the poor prediction by a single model. In this paper, a weighted averaging estimator of conditional mode is developed in modal regression, where the weights are selected by optimizing a kernel-based cross-validation criterion. The theoretical justifications are provided in two scenarios. First, when all the candidate models are misspecified, we derive that the resultant model averaging estimator is asymptotically optimal in the sense of minimizing the out-of-sample kernel-based prediction error. Second, when the candidate model set contains at least one correct model, we demonstrate that the estimated weights are asymptotically concentrated on the correct models. Additionally, to identify the key predictors in modal regression, a variable importance measure is developed based on our model averaging approach. The simulation study and real data analysis show that the proposed procedure has satisfactory prediction accuracy.</p> |
