A Differential Privacy Method for Publishing Regression Coefficients in Student Performance Analysis
As educational informatization deepens, data-driven learning analytics improve teaching quality while also raising concerns about the disclosure of sensitive student performance data. Focusing on individual inference risks in regression analysis and coefficient release for international student grades, this paper introduces and optimizes a budget allocation-based differentially private linear regression model (DPBA). Under strict data normalization, the method transforms the ordinary least-squares objective function into a quadratic form, allocates privacy budgets according to the different global sensitivities of its components, and injects Laplace noise. In this way, it balances privacy protection and data utility. An empirical study using real course data from international students shows that, compared with ordinary linear regression, the DPBA model substantially weakens the dominance of a single extreme sample over regression coefficient estimates and prediction outputs, and it effectively resists individual-data inference attacks. At the same time, prediction accuracy remains highly usable, with a coefficient of determination . This study provides education administrators with a reliable technical solution for safely releasing teaching statistics under strong privacy protection constraints.
[1] Romero C, Ventura S, 2020, Educational Data Mining and Learning Analytics: An Updated Survey. WIREs Data Mining and Knowledge Discovery, 10(3): e1355.
[2] Prinsloo P, Slade S, Khalil M, 2022, The Answer Is (Not Only) Technological: Considering Student Data Privacy in Learning Analytics. British Journal of Educational Technology, 53(4): 876–893.
[3] Mutimukwe C, Viberg O, Oberg LM, et al., 2022, Students’ Privacy Concerns in Learning Analytics: Model Development. British Journal of Educational Technology, 53(4): 932–951.
[4] Qu Z, Dai W, Euan C, et al., 2025, Exploratory Functional Data Analysis. Test, 34(2): 459–482.
[5] Jones KML, VanScoy A, Bright K, et al., 2022, A Measurement of Faculty Views on the Meaning and Value of Student Privacy. Journal of Computing in Higher Education, 34(3): 769–789.
[6] Schleiss J, Gunther K, Stober S, 2022, Protecting Student Data in ML Pipelines: An Overview of Privacy-Preserving ML. In Rodrigo MM, Matsuda N, Cristea AI, et al. (eds.), Artificial Intelligence in Education. Posters and Late Breaking Results, Workshops and Tutorials, Industry and Innovation Tracks, Practitioners’ and Doctoral Consortium. Cham: Springer International Publishing, 532–536.
[7] Yang Y, Zhang Z, Miklau G, et al., 2012, Differential Privacy in Data Publication and Analysis. In Proceedings of the 2012 ACM SIGMOD International Conference on Management of Data. Scottsdale, AZ: ACM, 601–606.
[8] Dwork C, McSherry F, Nissim K, et al., 2006, Calibrating Noise to Sensitivity in Private Data Analysis. In Theory of Cryptography Conference. Berlin: Springer, 265–284.
[9] Namatevs I, Sudars K, Nikulins A, et al., 2025, Privacy Auditing in Differential Private Machine Learning: The Current Trends. Applied Sciences, 15(2): 647.
[10] Fang X, Yu F, Yang G, et al., 2019, Regression Analysis with Differential Privacy Preserving. IEEE Access, 7: 129353–129361.
