Open AccessQuantification of Curcuminoids in Turmeric Using Visible Reflectance Spectra and a Decision-Tree Based Chemometric Approach
Author(s) -
Hasika Suresh,
Amruta Ranjan Behera,
Shankar Kumar Selvaraja,
Rudra Pratap
Publication year - 2020
Publication title -
journal of the electrochemical society
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.258
H-Index - 271
eISSN - 1945-7111
pISSN - 0013-4651
DOI - 10.1149/1945-7111/abd603
Subject(s) - decision tree , tree (set theory) , range (aeronautics) , computer science , chemometrics , curcumin , content (measure theory) , biological system , analytical chemistry (journal) , mathematics , materials science , chemistry , artificial intelligence , chromatography , machine learning , composite material , biology , biochemistry , mathematical analysis
For quantification of curcumin content in turmeric, a low-cost multivariate-analysis-based sensing system is desired. It can be realized by exploiting the spectra in the visible region, which enables the use of off-the-shelf, relatively inexpensive light sources and detectors. To address this, we propose a novel decision-tree method for improved prediction accuracy. Two sets of models with PLSR algorithm are developed with the measured reflectance spectra from 66 turmeric samples in the range of 360–750 nm, and their respective curcuminoids content are quantified by HPLC. A suite of a coarse-model for initial prediction of turmeric samples in the broad range of 1%–4%, and five finer-models for subsequent prediction (in the ranges 1%–2%, 2%–3%, 3%–4%, 1.5%–2.5%, and 2.5%–3.5%) constitute the proposed decision-tree approach. The method’s efficacy is substantiated from an improved coefficient of determination ( R 2 ) for the finer models (0.90–0.96) as compared to the coarse-model’s 0.92. This is further corroborated with lower RMSECV of 0.06–0.13 and an RMSEP of 0.15–0.25 for finer models, as compared to 0.219 and 0.45 for the coarse model, respectively. Testing reveals that the method results in 46% reduction in prediction error. Realization of a robust prediction approach in the visible range sets the stage for the development of cost-effective field-deployable devices for on-site measurement of curcumin.