Machine Learning–Driven Analysis of KOH and NaOH Performance in Biodiesel Production from Waste Frying Oil

This study aims to conduct a comparative multivariate analysis of potassium hydroxide (KOH) and sodium hydroxide (NaOH) catalysts for biodiesel production from waste frying oil using a multivariate analytical framework. A central composite rotatable design (CCRD) and response surface methodology (RSM) are integrated with machine learning (ML) techniques—random forest regression and principal component analysis (PCA)—to model, predict, and decode the complex interactions between process parameters. This proposed hybrid approach uncovers previously underexplored catalyst-specific mechanistic differences. The analysis indicates KOH's superior performance, achieving a higher mean biodiesel yield (90.7%) than NaOH (84.1%). Crucially, the ML models reveal a fundamental mechanistic divergence: the methanol-to-oil (MeOH) ratio is a dominant positive factor for KOH but a detrimental factor for NaOH. This catalyst-specific effect underpins distinct optimization strategies and confirms KOH as the essential catalyst for efficient biodiesel synthesis from waste frying oil.