Keywords: High Dimensional Data, Partial Least Squares Regression, Robust Multivariate Distances, RWSIMPLS, Weighted Likelihood
Spectrophotometric analytical methods, combined with chemometrical methods are often preferred for quantitative evaluation of drugs in pharmaceutical formulations. Signal overlapping, nonlinearity, multicollinearity and presence of uncalibrated outliers deteriorate the performance of analytical studies. The Partial Least Squares Regression (PLSR) is a very popular Chemometric method in quantification of high dimensional spectrally overlapped drug formulations. The SIMPLS is the mostly used PLSR algorithm, but it is highly sensitive to outliers that also effects the diagnostics. In this paper, we propose new robust multivariate diagnostics to identify outliers, influential observations and points causing non-normality. We study performances of the proposed diagnostics on two highly overlapping drug systems: Paracetamol-Caffeine and Doxylamine Succinate-Pyridoxine Hydrochloride.