Biodiesel fuel is known to reduce carbon footprint compared to fossil diesel and, thus, has a significant potential for compression ignition engine applications. Unlike diesel, biodiesel can be produced from various sources. Consequently, they vary in composition depending on the feedstock chosen, resulting in variability in their oxidative stability, engine durability, fuel spray and combustion characteristics. In the present study, machine learning-based predictive models were developed to predict biodiesel fuel properties and engine characteristics. The models were developed using multilinear regression (MLR), artificial neural networks (ANN), support vector machine regression with grid search (SVMGS), gaussian process regression (GPR), random forest (RF), and adaptive neuro-fuzzy inference system (ANFIS) algorithms. Among all the models, SVMGS models resulted in a lower mean absolute percentage error (MAPE) of less than 2% in predicting biodiesel viscosity, cetane number, and calorific value. Models developed using random forest regression best-predicted ignition delay, peak cylinder pressure, brake thermal efficiency, brake-specific fuel consumption, and engine exhaust emissions with a MAPE well below 4%. Further, the chosen models were used to determine the optimal biodiesel composition using the Genetic algorithm targeting lower viscosity, higher cetane number and calorific value. The optimization results suggest that biodiesel derived from a blend of camelina and coconut oils in volume percentages of 68%±1% and 32%±1% exhibits better fuel properties, improved fuel economy and lower exhaust emissions. Thus, the machine learning approaches emerge as a transformative ally in the exciting realm of biodiesel research to fine-tune the chemical profile of biodiesel feedstock for improved fuel properties and better engine characteristics.