Superviz25-SQL Dataset ANUBIS'2025

Dataset Description Superviz25-SQL is made to evaluate unsupervized SQL Injection detection techniques. The dataset allows to compare the capability of mechanisms trained on the benign workload of a specific database deployment to detect SQL Injection attack targeting this deployment. We provide a test and train split, the train set consist of 335306 benign queries. The test set comprise 3017390 normal samples and 335192 malicious samples (90:10). The dataset offers more than the SQL query and its label as extended metadata is provided to facilitate the comparison of detection approaches. For instance the attack technique information allows to compare the effectiveness of detection mechanism across existing SQL Injection techniques, and the user input field allows the comparison of approaches basing their detection on full queries to those solely using the user inputs. The code used for the generation of this dataset is available at gquetel/sqlia-dataset, and the associated paper at TODO. If you use this dataset, please acknowledge it by citing the original paper: TODO This work has been partially supported by the French National Research Agency under the France 2030 label (Superviz ANR-22-PECY-0008). The views reflected herein do not necessarily reflect the opinion of the French government. Usage The proposed training and testing sets can easily be loaded using python and pandas as follows: import pandas as pd df = pd.read_csv( "dataset.csv", dtype={ "full_query": str, "label": int, "user_inputs": str, "attack_stage": str, "tamper_method": str, "attack_status": str, "statement_type": str, "query_template_id": str, "attack_id": str, "attack_technique": str, "split": str, }, ) df_train = df[df["split"] == "train"] df_test = df[df["split"] == "test"]