DNA is now routinely used in criminal investigations and court cases, although DNA samples taken at crime scenes are of varying quality and therefore present challenging problems for their interpretation. We present a statistical model for the quantitative peak information obtained from an electropherogram of a forensic DNA sample and illustrate its potential use for the analysis of criminal cases. In contrast with most previously used methods, we directly model the peak height information and incorporate important artefacts that are associated with the production of the electropherogram. Our model has a number of unknown parameters, and we show that these can be estimated by the method of maximum likelihood in the presence of multiple unknown individuals contributing to the sample, and their approximate standard errors calculated; the computations exploit a Bayesian network representation of the model. A case example from a UK trial, as reported in the literature, is used to illustrate the efficacy and use of the model, both in finding likelihood ratios to quantify the strength of evidence, and in the deconvolution of mixtures for finding likely profiles of the individuals contributing to the sample. Our model is readily extended to simultaneous analysis of more than one mixture as illustrated in a case example. We show that the combination of evidence from several samples may give an evidential strength which is close to that of a single-source trace and thus modelling of peak height information provides a potentially very efficient mixture analysis.

Statistical analysis of DNA mixtures is known to pose computational challenges due to the enormous state space of possible DNA profiles. We propose a Bayesian network representation for genotypes, allowing computations to be performed locally involving only a few alleles at each step. In addition, we describe a general method for computing the expectation of a product of discrete random variables using auxiliary variables and probability propagation in a Bayesian network, which in combination with the genotype network allows efficient computation of the likelihood function and various other quantities relevant to the inference. Lastly, we introduce a set of diagnostic tools for assessing the adequacy of the model for describing a particular dataset.

No abstract

In Cowell et al. (2007a) a Bayesian network for analysis of mixed traces of DNA was presented using gamma distributions for modelling peak sizes in the electropherogram. It was demonstrated that the analysis was sensitive to the choice of a variance factor and hence this should be adapted to any new trace analysed. In the present paper we discuss how the variance parameter can be estimated by maximum likelihood to achieve this. The unknown proportions of DNA from each contributor can similarly be estimated by maximum likelihood jointly with the variance parameter. Furthermore we discuss how to incorporate prior knowledge about the parameters in a Bayesian analysis. The proposed estimation methods are illustrated through a few examples of applications for calculating evidential value in casework and for mixture deconvolution.

No abstract

In the pharmaceutical industry, the maintenance of production machines must be audited by the regulator. In this context, the problem of predictive maintenance is not when to maintain a machine, but what parts to maintain at a given point in time. The focus shifts from the entire machine to its component parts and prediction becomes a classification problem. In this paper, we focus on rolling-elements bearings and we propose a framework for predicting their degradation stages automatically. Our main contribution is a kmeans bearing lifetime segmentation method based on highfrequency bearing vibration signal embedded in a latent lowdimensional subspace using an AutoEncoder. Given highfrequency vibration data, our framework generates a labeled dataset that is used to train a supervised model for bearing degradation stage detection. Our experimental results, based on the FEMTO Bearing dataset, show that our framework is scalable and that it provides reliable and actionable predictions for a range of different bearings.