Logistic regression classification model identifying drugs of abuse based on their ATR-FTIR spectra: Case study on LASSO and Ridge regularization methods

Abstract

We are presenting a comparative study of two variable selection methods, i.e. Ridge and LASSO regularization methods that were used for selecting the most adequate coefficients for error minimization. The logistic regression model was used for the detection (classification) of new analogues of high risk drugs of abuse belonging to the 2C-x and DOx classes of hallucinogenic amphetamines, based on their ATR-FTIR spectra. The results indicate that the LASSO method yields the best results. Although both methods generate remarkably sensitive class identity recognition tools, the system based on the Ridge regularization method is less selective than that using the LASSO method.