What question did this study set out to answer?

This research aims to enhance the yield prediction and optimization processes for synthesizing ferulic acid using machine learning methods.

May 16, 2026

Machine Learning‐Driven Isolated Yield Prediction and Experimental Parameter Optimization for Ferulic Acid Synthesis

Key Points

This research aims to enhance the yield prediction and optimization processes for synthesizing ferulic acid using machine learning methods.
Constructed and evaluated ten machine learning algorithms for isolated yield prediction.
Employed Random Forest for fitting accuracy and generalization potential.
Utilized SHAP and Bi-PDPs to analyze the influence of individual parameters and their interactions.
The Random Forest model showed superior fitting accuracy compared to other algorithms.
Machine learning methods effectively predicted isolated yields, facilitating better optimization decisions.
SHAP and Bi-PDPs clarified the impact of various factors on the synthesis process.

Abstract

Chemical synthesis is an essential method for obtaining high‐purity ferulic acid (FA). However, due to the inherent complexity of the Knoevenagel condensation system and the nonlinear coupling among process parameters, traditional approaches often fail to achieve precise isolated yield prediction and optimization. This study introduces a machine learning (ML)‐guided experimental optimization strategy. Ten representative algorithms, including Gradient Boosting Decision Trees (GBDT), Random Forest (RF), Extreme Gradient Boosting (XGBoost), Support Vector Regression (SVR), Adaptive Boosting (AdaBoost), Categorical Boosting (CatBoost), k ‐Nearest Neighbors (KNN), Linear Regression, Decision Trees (DT), and Ridge, were constructed and systematically evaluated. The results indicate that the RF model exhibits superior fitting accuracy and exhibits potential for generalization. Based on the RF model, Shapley Additive exPlanations (SHAP) and Bi‐factor Partial Dependence Plots (Bi‐PDPs) were employed to quantitatively assess the influence of individual factors and to elucidate interactive effects between variables. This work establishes an efficient ML‐driven framework for the rational optimization of complex organic synthesis processes.

Mark Helpful

Bookmark

Relay