Recursive estimation of intensity function of a Poisson random field

Anscombe's condition asymptotic normality asymptotic sequential bounds asymptotic sequential fixed-width confidence interval asymptotic unbiasedness integrated squared error Poisson random field Ramlau-Hansen kernel intensity estimator recursive kernel estimator stopping rule strong consistency unknown intensity function

Mathematics Subject Classification ID

Density estimation (62G07) Sequential estimation (62L12) Optimal stopping in statistics (62L15) Non-Markovian processes: estimation (62M09) Random fields; image analysis (62M40)

Review text

Let \(N_ 1(z),\dots,N_ n(z)\), \(z\in[0,1]\times\cdots\times[0,1]=I\subset \mathbb{R}^ p\), be independent copies of a Poisson random field with unknown intensity function \(\lambda\) which is continuous on \(I\). Let \(K: \mathbb{R}^ p\to\mathbb{R}_ +\) be a bounded probability density function with support contained in \(Q=[- 1,1]\times\cdots\times[-1,1]\), \(K(z)=K(-z)\) for each \(z\in Q\), \(| z_ 1|\leq| z_ 2|\) implies \(K(z_ 2)\leq K(z_ 1)\), and the function \(z\to z^ 2K(z)\) is integrable. In order to estimate the intensity function \(\lambda(z)\), a recursive modification of the Ramlau-Hansen kernel intensity estimator is used: \[ \widehat\lambda(N_ 1,\dots,N_ n,z)=\widehat\lambda_ n(z)=n^{- 1}\sum^ n_{k=1}(b^ p_ k) ^{-1}\int_ IK((z-u)/b_ k))N_ k(du),\;b_ k=\alpha k^{-\beta}, \beta<1/p.\tag{1} \] Asymptotic unbiasedness, strong consistency and asymptotic normality of the recursive kernel estimator (1) of an intensity function of a Poisson random field are proved. Three classes of stopping times are defined, which permit constructing an asymptotic sequential fixed-width confidence interval for the intensity function \(\lambda\) at a fixed point and asymptotic sequential bounds based on the integrated squared error.

Recommendations

Cites work

Cited in

(3)

This page was built for publication: Recursive estimation of intensity function of a Poisson random field

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1205453)