This paper presents a machine learning–based framework for predicting overcrowding levels in public transport systems. The proposed system analyzes historical passenger flow data and key operational parameters to classify crowd levels into LOW, MEDIUM, and HIGH categories. Random Forest and XG Boost algorithms are employed as base classifiers, and their outputs are combined using an ensemble strategy to enhance prediction accuracy and robustness. The framework further integrates SHAP-based interpretability to explain model decisions and identify the most influential features affecting crowd levels. In addition, a real-time simulation module is incorporated to compare predicted and observed crowd conditions, enabling deviation detection and adaptive monitoring. Experimental evaluation demonstrates strong classification performance across all crowd categories, confirming the reliability and effectiveness of the ensemble approach. The proposed model supports intelligent transport management by providing accurate crowd predictions and interpretable insights, thereby assisting authorities in proactive decision-making, resource allocation, and passenger information systems. The overall system contributes to improving operational efficiency and enhancing commuter safety within urban public transportation networks