Improving the accuracy of estimation of unknown random variable probability density over empirical data (Q5944567)

An unknown probability density function can be estimated by the Parzen-Rosenblatt (PR) method. The authors construct from the PR estimates from two independent samples of size \(n\) by an extrapolation procedure an estimate which is more accurate than the PR estimate from the combined sample of size \(2n\). The questions of how to choose optimal parameters and how to construct confidence intervals, are investigated. The effectiveness of the method is demonstrated on an example.