Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Privacy-Preserving Credit Card Approval Using Homomorphic SVM: Toward Secure Inference in FinTech Applications

Faneela, Baraq Ghaleb, Jawad Ahmad, William J. Buchanan, Sana Ullah Jan

TL;DR

The paper tackles the challenge of performing credit-scoring classification in privacy-sensitive finance by enabling secure inference over encrypted data. It introduces PP-FinTech, a CKKS-based encrypted SVM using a hybrid Polynomial–RBF kernel and an adaptive threshold to maintain robustness under encryption noise, with an optimized SIMD-enabled inference pipeline. The approach trains the SVM in plaintext on an 80:20 split of the Credit Card Approval dataset and then encrypts the model to perform inference securely, achieving performance comparable to plaintext baselines while providing strong privacy guarantees. The results demonstrate practical latency (approximately 44.9 ms per sample) and scalable batch processing, highlighting the method's potential for secure fintech ML deployments.

Abstract

The growing use of machine learning in cloud environments raises critical concerns about data security and privacy, especially in finance. Fully Homomorphic Encryption (FHE) offers a solution by enabling computations on encrypted data, but its high computational cost limits practicality. In this paper, we propose PP-FinTech, a privacy-preserving scheme for financial applications that employs a CKKS-based encrypted soft-margin SVM, enhanced with a hybrid kernel for modeling non-linear patterns and an adaptive thresholding mechanism for robust encrypted classification. Experiments on the Credit Card Approval dataset demonstrate comparable performance to the plaintext models, highlighting PP-FinTech's ability to balance privacy, and efficiency in secure financial ML systems.

Privacy-Preserving Credit Card Approval Using Homomorphic SVM: Toward Secure Inference in FinTech Applications

TL;DR

The paper tackles the challenge of performing credit-scoring classification in privacy-sensitive finance by enabling secure inference over encrypted data. It introduces PP-FinTech, a CKKS-based encrypted SVM using a hybrid Polynomial–RBF kernel and an adaptive threshold to maintain robustness under encryption noise, with an optimized SIMD-enabled inference pipeline. The approach trains the SVM in plaintext on an 80:20 split of the Credit Card Approval dataset and then encrypts the model to perform inference securely, achieving performance comparable to plaintext baselines while providing strong privacy guarantees. The results demonstrate practical latency (approximately 44.9 ms per sample) and scalable batch processing, highlighting the method's potential for secure fintech ML deployments.

Abstract

The growing use of machine learning in cloud environments raises critical concerns about data security and privacy, especially in finance. Fully Homomorphic Encryption (FHE) offers a solution by enabling computations on encrypted data, but its high computational cost limits practicality. In this paper, we propose PP-FinTech, a privacy-preserving scheme for financial applications that employs a CKKS-based encrypted soft-margin SVM, enhanced with a hybrid kernel for modeling non-linear patterns and an adaptive thresholding mechanism for robust encrypted classification. Experiments on the Credit Card Approval dataset demonstrate comparable performance to the plaintext models, highlighting PP-FinTech's ability to balance privacy, and efficiency in secure financial ML systems.
Paper Structure (21 sections, 16 equations, 4 figures, 1 table, 2 algorithms)

This paper contains 21 sections, 16 equations, 4 figures, 1 table, 2 algorithms.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Support Vector Machine (SVM)
  4. CKKS Scheme
  5. Related Work
  6. Proposed Scheme
  7. Model Training and Feature Selection
  8. Hybrid Kernel Approach
  9. Conservative Scaling for Error Recovery
  10. Encrypted Inference
  11. Data Encryption and HE Kernel Evaluation
  12. SIMD-Based Encrypted Matrix Multiplication
  13. Adaptive Threshold for Secure Classification
  14. Secure Decryption and Classification
  15. Results and Discussion
  16. ...and 6 more sections

Figures (4)

  • Figure 1: Performance Comparison Between PT-Fintech and PP-FinTech
  • Figure 2: ROC Curve Comparison
  • Figure 3: Empirical analysis of runtime and noise behavior during encrypted inference.
  • Figure 4: Simulated scalability: Total inference time across different batch sizes for the PP-FinTech model.