Development of a Multiple Linear Regression Model for Predicting Key Fuel Properties of Jatropha curcas and Azadirachta indica-Based Biodiesel

Abstract

Accurate prediction of biodiesel fuel properties is essential for reducing experimental cost and improving the efficiency of fuel formulation. This study developed Multiple Linear Regression (MLR) models to predict the kinematic viscosity and calorific value of biodiesel–diesel blends produced from pure and hybrid Jatropha curcas and Azadirachta indica feedstocks. Biodiesel was produced from five feedstock formulations comprising pure Jatropha curcas, pure Azadirachta indica, and three intermediate hybrid mixtures, after which each biodiesel was blended with petroleum diesel at B5–B30 ratios. Experimental measurements of the selected fuel properties were used to establish regression relationships based on two predictor variables: Jatropha feedstock composition and biodiesel blend ratio. The developed models showed reasonable predictive capability, with coefficients of determination (R²) of 0.9860 for kinematic viscosity and 0.9948 for calorific value. Prediction errors were consistently low, with RMSE values of 0.0505 mm² s⁻¹ and 0.0446 MJ kg⁻¹, and MAE values of 0.0413 mm² s⁻¹ and 0.0356 MJ kg⁻¹ for kinematic viscosity and calorific value, respectively. Leave-One-Out Cross-Validation further confirmed good model performance, with cross-validated R² values of 0.9810 for kinematic viscosity and 0.9932 for calorific value. Residual diagnostics showed constant error variance for both models, while normality was better satisfied for the calorific-value model than for the kinematic-viscosity model. The findings show that simple MLR equations can provide useful estimates of key fuel properties from feedstock composition and biodiesel blend ratio within the experimental range investigated, thereby reducing the need for repeated laboratory analyses. The proposed models therefore provide a practical and computationally efficient approach for biodiesel formulation, fuel quality evaluation and preliminary engineering analysis, while offering a useful basis for future predictive studies involving other non-edible biodiesel feedstocks.