safeboost1/SAFEboost: SAFEBOOST

SAFE-Boost: Privacy-Preserving XGBoost using Homomorphic Encryption SAFE-Boost is a framework that performs XGBoost training and inference while preserving data privacy using Homomorphic Encryption (HE). This project is built upon the HEaaN.Stat library. Installation & Setup Prerequisite: HEaaN.Stat (Docker) This project requires the HEaaN.Stat environment for encrypted computations. Please visit the link below to set up the environment in your Docker container: HEaaN.Stat Installation: https://iheaan.com/?target=heaan-stat Clone the Repository Clone this repository inside your configured Docker environment: git clone https://github.com/safeboost1/SAFEboost.git cd SAFEboost Quick Start: Training & Inference Run run.py in the SAFE_Boost directory to perform the entire process from training to inference under homomorphic encryption. cd SAFE_Boost pip install -r requirements.txt # Install required dependencies python run.py To change the dataset or adjust hyperparameters, modify the following configuration in SAFE_Boost/run.py dataset_name = "iris" # Supported: "iris", "default_of_credit_card", "breast-cancer" max_bin = 8 # Max bins (Choose from [4, 6, 8, 10, 12]) depth = 3 # Tree depth ntree = 3 # Number of trees If you wish to use other datasets, please reach out via the email address listed in the Contact section. Functionality: Executes HE-based Training and Inference. Output: You can verify the accuracy of the encrypted computation results alongside the plaintext model for comparison. Benchmark: SAFE-I vs Levelup (RCC) The benchmark directory allows you to compare SAFE-I (SAFE-Boost's Inference) and Levelup’s RCC (Reduced Complexity Comparison) methods in an encrypted environment. 1. Pre-step: Generate Tree Before starting the benchmark, you must generate the tree models: python generate_tree.py Running Benchmarks Navigate to each folder (SAFE-I, levelup) and run the benchmarks for the Iris, Breast-Cancer, and Spambase datasets. Run SAFE-I: cd benchmark/SAFE-I && python run.py Run Levelup: cd benchmark/levelup && python run.py All results (performance metrics and logs) are saved in the exp directory within each respective folder. Troubleshooting: Key Generation If you encounter the following error: heaan_stat.exceptions.SDKHEError: [HEaaN.Stat SDK] Secret key does not exist Please modify the generate_keys argument in the Context configuration: If keys do not exist (First run): Set generate_keys=True If keys already exist: Keep generate_keys=False context = Context( parameter=param, key_dir_path='./keys', generate_keys=True, # Change to True if keys are missing ) Execution Modes: Simulation vs. Real HE Since Homomorphic Encryption (HE) is computationally intensive and this version of the library is CPU-bound, we provide a Simulation Mode for logic verification. You can toggle this via an environment variable in run.py. Simulation Mode (pi): This is the default setting. It mimics HE logic using plaintext, allowing you to verify the algorithm flow quickly. Real HE Mode (real): This performs actual encrypted computations. It is significantly slower and requires more memory/CPU resources. import os os.environ["HEAAN_TYPE"] = "pi" # Change to "real" for actual encryption Notice (Commercial Use & Performance) Public Library Version: Due to commercial security policies and licensing restrictions, this repository contains the public version of the library. Performance Note: Please note that the performance may differ from the results reported in official papers or commercial environments depending on the library version and hardware configuration used. Contact For inquiries regarding performance on additional datasets, detailed implementation details, or collaboration, please contact us via email. Email: safeboost1@gmail.com