The prediction of biodiesel production yield from transesterification of vegetable oils with machine learning

• The yield of biodiesel production from various vegetable oils and catalysts were predicted by machine learning. • The ANN and ANN-PSO were applied for model biodiesel yield in terms of maximum r square and minimum mean absolute percentage error. • For the combination dataset between alkali catalyzed process and acid catalyzed process, the ANN with trainbr algorithm performed the most accuracy. Biodiesel is a renewable energy produced from transesterification of vegetable oils in the presence of various catalysts. The biodiesel production yield depends on types of feedstocks, reaction time, temperature, ratio of methanol to oil, and types of catalysts. Machine learning can be applied for prediction of biodiesel yield in a single process but the combination of biodiesel production processes, including a variety of alkali and acid catalysts, should be investigated. This work developed the modeling by using machine learning to predict the biodiesel yield of alkali catalysts and acid catalysts. The machine learning algorithms consisted of artificial neural network (ANN) and artificial neural network – particle swarm optimization (ANN-PSO). The proposed model was examined for Case I (alkali process) and Case II (combination process). For Case I with 19 input variables and a single output variable, the ANN with the trainlm algorithm and the Tansig- Tansig activation function was the most suitable model, offering the MAPE of 1.3510 and R square of 0.99272. While in Case II, the ANN with the trainbr algorithm performed the most accurately in terms of R square of 0.97229 and MAPE of 2.32145. Finally, the feature importance for Case I and Case II presented that the molar ratio of methanol to oil was the most important variable for the prediction of biodiesel yield.