Sample Dataset for Automated Detection of Cheating Codes

This project includes a resampled dataset and statistical code for analyzing various subcategories of cheating codes. The dataset is derived from an initial collection of SQL queries from HackerRank (https://zenodo.org/record/8199741). Originally, the dataset contained 5,767,890 correct codes and only 1,992 cheating codes. To address the issue of data imbalance, we implemented a resampling strategy to achieve a more balanced data distribution. Specifically, undersampling techniques were used to reduce the number of correct codes while retaining all cheating codes. From 22 programming problems, we randomly selected 20,000 correct codes from each problem. This approach aimed to balance data representation across different problem types, enhancing the model’s adaptability and generalization. After resampling, the dataset consisted of 440,000 correct codes and 1,992 cheating codes.