Output details

Bayesian linear regression and variable selection for spectroscopic calibration

The research reported in this paper proposes and validates a new approach to spectroscopic calibration modelling. It simultaneously enables the achievement of i) a balance between model accuracy and complexity; (ii) providing a predictive distribution that automatically gives prediction intervals; (iii) removing the computational burden of cross-validation; and (iv) realising a variable selection strategy, a combination of characteristics which previous models had not provided. The work was carried out as part of a DTI Succeeding through Innovation Award (Implementing PAT for successful development of Biopharmaceuticals in the UK), led by GlaxoSmithKline.