A Proposed Framework for Estimating Missing Values in Biofuel Feedstock Selection

Biofuel production is becoming a necessary activity due to rising fossil fuel costs, shrinking oil reserves, rising GHG (particularly CO2) emissions that lead to global warming., This is particularly important for the transport sector. The viability of a feedstock for biodiesel production can be determined via data-driven modelling. For most prediction systems, missing data is a major concern. List-wise deletion, mean/median imputation, random imputation, K-nearest neighbour, clustering, and multiple imputations are some of the techniques available. Missing data imputation has improved classifier performance in many studies. However, this requires training, and which is time-consuming. In this study, we propose a framework where a Multilinear Regression combined with Cross-Validation and Feature Selection is applied to estimate missing fatty acid composition parameters. A database comprising of 332 feedstocks with 8 properties is used to construct an imputation algorithm. The coefficient of determination (R
²
) is used as the evaluation metric. An outcome is a tool that reads a file containing feedstocks and their features and generates a file with the missing values imputed. The results show the significance of modern data science methods in data imputation for biodiesel parameters enabling reliable and least biased comparison of biodiesel feedstocks.