Enhancing QSAR Modeling: A Fusion of Sequential Feature Selection and Support Vector Machine

Quantitative Structure-Activity Relationship (QSAR) modeling is an approach employed to predict the biological response of chemical compounds by considering their structural attributes. Classification machine learning algorithms can learn patterns and relationships between chemical structure (descriptors) and biological activity from datasets and then use this knowledge to predict active or inactive compounds. This study introduces a new approach that combines Sequential Feature Selection (SFS) with the Support Vector Machine (SVM) algorithm to select the most relevant molecular descriptors for QSAR modeling. SFS and SVM work collaboratively to identify the best subset of descriptors, resulting in improved predictive accuracy. The key steps include selecting an appropriate subset of descriptors using SFS from a larger set, SVM models are built using different subsets of descriptors, and the most accurate model is chosen for final use. As shown by measuring Accuracy and precision, In our study, the SFS-SVM model demonstrated superior performance in QSAR modeling on both DKPES and PubChem datasets. It achieved the highest accuracy and precision among the compared models, with 92.87% accuracy and 86.66% precision on DKPES, and 98.18% accuracy and 98.00% precision on PubChem. This outperformed other models like SFS-LR, Cluster-GA, RF, DT, and SVM, highlighting its efficiency in feature selection and predictive accuracy.