A systematic review and meta-analysis evaluated the effects of suspension-assisted training (SST) on dynamic balance in stroke patients based on 12 randomized controlled trials with 584 patients. SST significantly improved dynamic balance with an effect of SMD = 0.87 (95% CI 0.49–1.26, p < 0.0001), with heterogeneity I² = 76.6%. The overall quality of the evidence was moderate according to GRADE. Greater effects were seen with training lasting 6–8 weeks, sessions of 40–150 minutes, in patients in the subacute phase, combined approaches, and supporting approximately 30% of body weight. XGBoost-SHAP analysis showed that training duration and intervention mode were the most influential factors. Preliminary dose–response modeling suggested an optimal regimen of 90–100 min per session, 6 sessions per week for 8–9 weeks (total 34.6–37.6 h, approximately 36 h). These recommendations are considered hypothesis generating due to the heterogeneity and limitations of the studies.