This study proposes a deep learning-based approach to classify loops in programming code based on their parallelizability. Using two genetic algorithm-based code generators, we generated two types of code: (i) independent loops that are parallelizable and (ii) ambiguous loops whose dependencies are unclear, making it difficult to determine whether they are parallelizable. The generated code fragments were tokenized and preprocessed to create a robust dataset. Two deep learning models, a deep neural network (DNN) and a convolutional neural network (CNN), were implemented to perform classification. Robust statistical analysis was performed to validate the predicted performance of the DNN and CNN models across 30 independent runs. While the CNN exhibited slightly higher average performance, the two models exhibited similar variability. Experiments with various dataset sizes highlighted the importance of data diversity on model performance. These results demonstrate the feasibility of automatically identifying parallelizable structures in code using deep learning, offering a promising tool for software optimization and performance enhancement.
