Quantitative structure-property relationship modelling for predicting retention indices of essential oils based on an improved horse herd optimization algorithm.

The researchers have made advancements in predicting the retention indices of essential oils using an improved version of the Horse Herd Optimization Algorithm (HOA), which is a modern metaheuristic algorithm. They tested Z-shape transfer functions (ZTF) as part of the updated algorithm, with given results showing definitive improvements in prediction performance and faster convergence behaviour than other binary algorithms.

Understanding the Improved Horse Herd Optimization Algorithm (HOA)

• HOA is a relatively new metaheuristic algorithm, originally inspired by the social behavior of horse herds. It is particularly noted for its effectiveness in complex optimization tasks.
• In this study, researchers used a binary form of HOA (BHOA) to address the descriptor selection issue associated with predicting essential oil retention indices. This means that they used a version of HOA specifically designed to perform optimization in binary systems, which simplifies the selection of variable properties, or descriptors, that influence the retention indices of various essential oils.

Z-Shaped Transfer Functions (ZTF)

• ZTF were introduced in this study to analyze their efficiency in enhancing the BHOA’s performance for quantitative structure-property relationship (QSPR) modelling.
• The effect of the ZTF, especially ZTF1, in enhancing BHOA’s performance was measured using both internal and external prediction criteria. This includes the mean-squared error (MSE) of the training and testing datasets, leave-one-out internal and external validation, and the degree of convergence of the Z-shaped transfer functions.

Findings of the Research

• The primary finding from the study was that ZTF, especially ZTF1, significantly improved the performance of the original BHOA.
• ZTF1 was found to have the quickest convergence behaviour amongst all binary algorithms tested which creates a more efficient predicting model as it needs less iterations to reach a solution.
• ZTF1 also uses the least amount of descriptors, which means it simplifies the model without sacrificing prediction accuracy.

Research Methodology

• One important validation method used in the study was K-fold cross-validation, specifically with k=5, which helps estimate machine learning model performance on an independent dataset.