In the field of big data analytics, many challenges emerge from the high dimensionality of datasets. This study introduces an approach for feature selection that efficiently facilitates large-scale data processing by integrating machine learning with a heuristic optimization algorithm. The approach employs a Support Vector Machine (SVM) algorithm alongside the Black Widow Optimization (BWO) metaheuristic algorithm to efficiently handle high-dimensional datasets. It also explores integrating a K-Nearest Neighbor (KNN) algorithm with BWO, using various classification metrics different from what we used for evaluation in previous works. The empirical assessment consists of two experimental sets. The first set compares the performance of binary BWO with SVM (BBWO-SVM) against binary BWO with KNN (BBWO-KNN). The second set evaluates the performance of BBWO-SVM and BBWO-KNN against six other metaheuristic algorithms, using 28 public datasets from the UCI repository. The experimental results suggest that incorporating BWO with SVM leads to better performance compared to established metaheuristic algorithms, enhancing classification across diverse datasets. Moreover, the findings show that BWO-SVM performs comparably to BWO-KNN and exhibits slight improvements in certain cases. These results suggest that BBWO-SVM offers a noteworthy combination of accuracy and reliability, positioning it as a valuable tool in critical fields such as NLP, finance, and cybersecurity. The outcomes can potentially impact how complex data is analyzed and applied in these important sectors.