Direct and simultaneous determination of four phenolic antioxidants in biodiesel using differential pulse voltammetry assisted by artificial neural networks and variable selection by decision trees

Abstract A new methodology using differential pulse voltammetry and artificial neural network (ANN) for simultaneous determination of butylated hydroxyanisole (BHA), butylated hydroxytoluene (BHT), propyl gallate (PG) and tert -butylhydroquinone (TBHQ) in biodiesel samples is proposed. A platinum ultramicroelectrode (ume) was used as working electrode and measures were taken directly in biodiesel:ethanol medium without previous preparation. On this condition, detection limits for the antioxidants separately are 20.5, 32.4, 35.5 and 26.5 mg L −1 for BHA, BHT, PG and TBHQ, respectively. The artificial neural network model allowed the quantification of the individual concentrations overcoming the strongly overlapped voltammograms obtained for the mixture of the four antioxidants. For the model construction, a variable selection step through decision trees (DT) led to a reduction of prediction errors by 17.5%. The optimized DT-ANN model presented high correlation (0.97474, 0.99995, 0.98246 and 0.98928 for BHA, BHT, PG and TBHQ, respectively) between real and predicted values. Recovery percentages found were between 82.6% and 106.7%, except for two samples whose values were 76.0% and 114.7%. From the accuracy found between nominal and estimated concentration, it is inferred that the proposed methodology is a good alternative to quantify phenolic antioxidants in biodiesel samples.