With the accelerated digital transformation of urban governance, machine learning’s wide application in community planning and governance is reshaping traditional spatial governance. Yet, while algorithm-driven governance boosts efficiency, its inherent logic may exacerbate spatial inequality and trigger environmental justice issues. Based on environmental justice theory, this study develops a "cognitive-procedural-distributive justice" analytical framework. Via systematic literature review, it quantitatively analyzes 1996–2025 domestic and international studies, examining machine learning’s justice risks and reconstruction paths in community governance. The findings are as follows: (1) Cognitively, biased data and platform monopoly exclude vulnerable groups’ knowledge; (2) Procedurally, algorithmic black boxes and formal participation weaken planning democracy; (3) Distributively, efficiency-oriented optimization reinforces existing spatial inequality. Accordingly, countermeasures—including a community knowledge sovereignty mechanism, participatory algorithmic governance, and fairness-oriented evaluation system—are proposed to guide community planning’s shift from technical to justice rationality.