Predicting startup success remains a complex and high-stakes challenge in entrepreneurship and investment analysis. This study introduces an Optimized Bagging Tree Model enhanced through Bayesian Hyperparameter Tuning to improve classification accuracy in imbalanced startup datasets. A structured machine learning pipeline—comprising data preprocessing, feature selection, and model training—was applied to evaluate the model against standard classifiers such as Decision Trees, Support Vector Machines (SVMs), and Neural Networks. The proposed approach achieved a notable accuracy improvement, increasing from 64.7% to 85.9%, particularly in identifying failed startups, which are often misclassified due to class imbalance. Feature analysis revealed that factors such as investment relationships, funding milestones, and business network characteristics significantly influence startup outcomes. While the proposed model demonstrated superior predictive performance, the study acknowledges limitations related to dataset imbalance and the exclusion of broader economic and policy variables. Future research is encouraged to explore advanced deep learning methods and integrate macroeconomic indicators to enhance model generalizability. These findings support the development of more reliable, data-driven decision-making tools for investors, entrepreneurs, and policymakers operating in uncertain, high-risk environments.