Statistical Discriminant Analysis in Forensic Science

A recent paper in the Journal of the Forensic Science Society [1] described a well-known statistical method for addressing the problem of discrimination of cat and dog hairs. It used a so-called “parametric” method which assumes that the observations are Normally distributed and estimates the parameters of the distribution from the data. The present paper describes another method of statistical discrimination known as the kernel method which dispenses with the assumption of Normality. The data alone determine the form of the distribution of the data and therefore multimodal or skew distributions may be more accurately modelled. The kernel method is applied to the data given in [1] and the results obtained compare favourably with those given in that paper.