Fire is a global hazard that societies worldwide face. According to the U.S. Fire Administration, home fires in the U.S. resulted in 2,070 fatalities in 2025 alone, many of which could have been prevented. This highlights the critical need for an evacuation simulation framework that is accessible to the public. We established a framework incorporating human-like evacuation agents with hierarchical path-finding algorithms. Moreover, Sense Fusion and Sector-Based Information Sharing Approach are introduced to mimic human perception and communication, respectively. Two machine learning models are built upon this framework, one of which is a surrogate model aiming to imitate the framework output, and the other is a Siamese network trained with pairwise loss. The former indicates that certain patterns can be identified and learnt by the model, while the latter enables the model to possess a higher degree of interpretability and allows researchers to directly compare floor plans in a pairwise manner. Multiple simulations were conducted to determine the impact of several variables on the results. The simulation results indicate that the proposed framework is both reliable and credible, whereas the two machine learning models demonstrate its identifiability and interpretability. Beyond technical performance, the proposed framework provides the public with an accessible and scalable means to analyze and evaluate evacuation routes, thereby improving public safety.