Estimation of Conditional Density Functions by Conformal Prediction and Model Averaging

In this research, we put forward a type of model averaging methods for estimating the conditional density function, based on a non-parametric estimation method and two different loss functions. Such methods provide accurate and stable estimation of the conditional density function. In addition, we develop prediction bands for the conditional density function in the case of finite samples by combining conformal prediction and model averaging. Conclusions from computational simulations and real data assessments based on photometric redshift estimation indicate the superiority of our proposed methods in comparison to other alternative methods.