Modeling and optimization of non-edible papaya seed waste oil synthesis using data mining approaches

Fossil fuels are a major contributor of greenhouse gas emissions (CO₂, NO_x, etc.). These fuels are non-renewable energy sources that will eventually be exhausted. Currently, biodiesel has gained attention as a renewable green energy source and means of supporting the minimization of fossil fuel use. However, obtaining biodiesel from edible seeds has been criticized as unethical, due to the source material being needed for human consumption. In this regard, papaya seed waste could be utilized as a potential feedstock because it is non-edible and its high lipid content makes it excellent for producing biodiesel. Papaya fruit is not seasonal. It is available at all times in tropical countries, and almost 75% of the total papaya is generated in the glove. Thus, different soft computing or data mining approaches such response surface methodology (RSM), artificial neural networks (ANNs), and support vector regression (SVR) can be utilized to predict oil yields from waste papaya seeds via solvent extraction. In the present research, the data for oil yields were obtained by experiments based on a central composite design. These data were then employed to develop, compare, and assess the suggested models. The results indicate that the SVR model performed much better for predicting oil yields than did the ANN and RSM models, with respect to various performance-measuring parameters (i.e., relative error, correlation coefficient, mean absolute error, and root mean squared error). It was observed that oil yields increase with an increase in extraction time but decrease as particle size increases. In order to find the global optimal set, an SVR and crow search algorithm-based interface was implemented. A maximum oil yield of 28.55% was achieved at 6.5 h of extraction and a particle size of 0.85 mm. The predicted oil yield was validated experimentally with less than a 5% rate of error. The extracted oil was also characterized by gas chromatography–mass spectrometry analysis.