Dependence of composition-based approaches on hybrid biodiesel fuel properties prediction using artificial neural network and random tree algorithms

Hybrid biodiesel (HB) synthesized via mixing two or more oils showed improved fuel properties and engine performance among other advantages and it is presently receiving attention in the research community. Different fatty acid methyl ester (FAME) composition-based approaches have been developed to improve the prediction of biodiesel fuel properties from the FAME constituents. However, for HB, the effect of these approaches on the fuel properties is lacking in the literature. This paper investigated the performance of the artificial neural network (ANN) and random tree to predict HB fuel properties from three FAME composition-based approaches (long chain saturation factor and degree of unsaturation (LCSF-DU), modified FAME compositions (MFC), and straight chain saturation factor and modified degree of unsaturation (SCSF-DUm)). FAMEs and fuel properties of HB sourced from the literature were used to develop ANN and random tree (RT) models. Data from the FAME composition-based approaches and fuel properties were used as input and output parameters respectively to develop these models. Results showed that the RT outperformed ANN in predicting the fuel properties for all the FAME composition-based approaches as marked by slightly higher R2 (ANN = 0.9921–0.9992 and RT = 0.9999–1.0000) and lower prediction errors (ANN (MAPE = 0.2684–8.0921 and RMSE = 0.0221–1.9732) and RT (MAPE = 0.00001–0.1691 and RMSE = 0.00001–0.0368). The R2 values demonstrated excellent performance of the developed ANN and RT models in predicting the fuel properties using different FAME composition-based approaches. Low prediction errors between the predicted and experimental-derived values of the fuel properties showed near experimental values prediction of these properties. With ANN, the use of the MFC approach was best in modeling the fuel properties whereas the RT modeling was independent of the FAME composition-based approaches. Calorific value and kinematic viscosity were best predicted using the SCSF-DUm approach while oxidative stability and flash point were predicted most using the LCSF-DU approach. Both the FAME composition-based approaches and learning algorithms have been revealed to influence the prediction of HB fuel properties.