Optimization based comparative study of machine learning methods for the prediction of bio-oil produced from microalgae via pyrolysis

Prediction of bio-oil yield using machine learning methods is an effective and economical approach. To examine the correlation between pyrolysis conditions, ultimate, and proximate analysis with bio-oil production is intricate and challenging task for the experimental techniques. Therefore, an efficient and well-organized model must be created to reliably predict the effect of input parameters on the bio-oil yield. Multiple ML models are integrated with PSO and GA for selection of features and hyperparameters optimization. Here, GPR-GA model performed better with R2 = 0.997 and RMSE = 0.0185 as compared to GPR-PSO model with R2 = 0.994 and RMSE = 0.0120. The values of R2 for DT, ANN, ET, SVM integrated with PSO are respectively 0.91, 0.92, 0.83 and 0.43 and for GA based algorithms the values of R2 are 0.62, 0.93, 0.94, and 0.55 respectively. The significance of input factors on bio-oil yield was thoroughly examined using partial dependence plots and Shapley method. Moreover, an interface was developed by GPR-GA model to predict bio-oil yield. Yield comparison predicted by GPR-GA model and experimental study shows remarkable synchronization with maximum error of 1.48. It offers tech-nologically advanced process in the pyrolysis of microalgae to enhance production of bio-oil.

Automated summary

Prediction of bio-oil yield using machine learning methods is effective and economical. The correlation between pyrolysis conditions, ultimate, and proximate analysis with bio-oil production is complex and challenging. Multiple ML models integrated with PSO and GA were used to predict the effect of input parameters on bio-oil yield.