Density, Kinematic Viscosity and Cetane Number Prediction of Biofuel Feedstocks from Fatty Acid Compositions

Stringent emission laws and the depletion of fossil fuels have made biodiesel one of the promising renewable and environmentally friendly fuel alternatives of the future. To get dependable engine performance and safe emission characteristics, it is imperative to know the Cetane number, kinematics viscosity and density of any fuel. While it is experimentally expensive and requires a continued concerted effort to know the actual properties of a fuel feedstock, recent Machine Learning methods offer wide possibilities to predict with close proximity certain characteristics of biomaterials within well-defined confidence levels. The present paper uses six different machine learning algorithms to predict three basic properties of fifty different biodiesel feedstocks from their fatty acid compositions. The issue of sparsity and anomaly in the data is dealt with through tested methods, following which properties for each feedstock are predicted within good accuracy levels. It is observed that the percentage error is very closely related to the coefficient of variation of a property, which will allow scientists to set variability bounds of predicted values.