Aiming at the problems of frequent model performance degradation, experience-dependent repair strategies, and lack of quantitative evaluation in machine learning applications, this study constructs a causal inference-driven evaluation and personalized decision framework for model repair based on large-scale controlled data. The average treatment effects are quantified through PSM, IPW, and AIPW, while the heterogeneous treatment effects are estimated by combining causal forests. Effective repair patterns are identified by integrating feature and cost-benefit analysis. This research upgrades model repair from empirical decision-making to data-driven scientific decision-making, provides implementable repair guidelines for industry, and offers a new paradigm for the application of causal inference in machine learning operations and maintenance.

Dong Jiaqi. Quantification of Validity and Personalized Decision Framework for Machine Learning Model Repair Strategies Based on Causal Inference. Academic Journal of Computing & Information Science (2026), Vol. 9, Issue 3: 11-16. https://doi.org/10.25236/AJCIS.2026.090302.

[1] Yanagaki S ,Yoshida S ,Oguro S , et al. Cost-effectiveness analysis of a bone metastasis cancer board for skeletal-related events of breast cancer[J].Japanese journal of clinical oncology,2026.

[2] Woldemariam A A ,Zhou Y ,Qiu Y .A novel average treatment effect estimation approach for integrated incomplete datasets, with application in childhood anemia[J].BMC medical research methodology,2026.

[3] Xie M ,Geldsetzer P ,Blakeman T , et al. A population based, regression discontinuity analysis examined the effects of nationwide alerting for acute kidney injury on healthcare and patient outcomes[J].Kidney international,2026.

[4] Rice M W ,Shaw P A ,Britt C R , et al. What's the Rush? Challenging the Early Surgery Paradigm for Older Adults in Emergency General Surgery[J].Journal of the American College of Surgeons, 2026.

[5] Han R ,Guo Y ,Zhang M , et al. Dose-response relationship between early pregnancy blood pressure and gestational diabetes mellitus based on propensity score matching: a retrospective study[J]. BMC pregnancy and childbirth,2026.

[6] Zhang J R ,Chen Z R ,Wang T X , et al. Radiation-induced hepatic toxicity from radiotherapy plus immune checkpoint inhibitors with targeted therapy in HCC patients: a propensity score-matched study[J].Radiation oncology (London, England),2026.

[7] Agbi M D ,Mwebesa E ,Jimmy I A , et al. Impact of eight or more antenatal care visits on intermittent preventive treatment of malaria uptake during pregnancy and facility-based delivery in Ghana: a propensity score matched analysis[J].BMC pregnancy and childbirth,2026.

[8] Dong Y C ,Wang X ,Zhu Y M , et al. Single vs. double good-quality embryo transfer in fresh cleavage-stage cycles: a propensity score-matched analysis of efficacy and risks[J]. Journal of ovarian research,2026.

[9] Heidari A ,Ebrahimi A ,Mirkhalafzadeh M , et al. Association between prenatal exposure to residential pyrethroid insecticides and congenital hypothyroidism using propensity score matching[J]. Scientific reports,2026.

[10] Baltussen C J ,Glas D A N ,Cessie L S , et al. How to Analyze Longitudinal Patient-Reported Outcomes in Populations With High Mortality Rates[J]. Journal of the American Geriatrics Society, 2026.