This study presents a simulation-based optimization framework to determine the optimal number of spare trainsets for high-frequency urban railways under failure uncertainty. Monte Carlo simulations replicate daily operations with stochastic failures, generating KPIs such as average delay and net profit across different spare allocations. Results show a clear optimal spare count that maximizes profit while maintaining punctuality, with robustness confirmed across varying failure rates. A regression-based estimator is also developed to predict the optimal spare requirement directly from fleet size and failure rate. This enables rapid, simulation-free planning for practical railway fleet management.