The rapid progress in technology and information sciences, along with the increasing demand for automation, has created a growing need for code that defines operations and tasks. Software projects typically depend on dedicated testing teams and extensive procedures to prepare and implement processes into a product. Although crucial, software testing is often underestimated in developing reliable software. Additionally, developers are increasingly adopting code generators to reduce development time. This study investigates the potential of combining natural language processing (NLP) with classifier algorithms to detect software defects in Python source code. A semi-synthetic dataset is employed, and a classifier is used for fault detection. Due to the high dependency of classifiers on parameter selection, optimizers are necessary to achieve optimal performance. This research introduces a modified version of the VNS algorithm to address optimization requirements. Findings suggest that NLP for code error detection is a promising approach, with the top models reaching an accuracy of over 99.7%.