Feature selection is a crucial preprocessing step in data mining and machine learning, enhancing model performance and computational efciency. This paper investigates the efectiveness of the Side-Blotched Lizard Optimization Algorithm (SBLA) for feature selection by developing six novel variants: Sbla-s1, Sbla-s2, Sbla-s3, Sbla-v1, Sbla-v2, and Sbla-v3, each employing distinct S-shaped or V-shaped transfer functions to convert the continuous search space to a binary format. These variants were rigorously evaluated on nineteen benchmark datasets from the UCI repository, comparing their performance based on average classifcation accuracy, average number of selected features, and average ftness value. The results demonstrated the superiority of Sbla-s3, achieving an average classifca- tion accuracy of 92.8% across all datasets, a mean number of selected features of 20, and an average ftness value of 0.08. Furthermore, Sbla-s3 consistently outperformed six other state-of-the-art metaheuristic algorithms, achieving the highest average accuracy on six- teen out of nineteen datasets. These fndings establish Sbla-s3 as a promising and efective approach for feature selection, capable of identifying relevant features while maintaining high classifcation accuracy, potentially leading to improved model performance in various machine learning applications.