This project investigates how data science methods—particularly random forest regression and logistic regression—indicate key factors influencing robotic grasp performance in a simulated environment. Using NVIDIA Isaac Sim-Lab and a Franka Emika: Panda arm manipulator, 1,500 synthetic episodes were collected capturing (sub) reward metrics (e.g., reaching, lifting, action-rate) and error measurements (positional, orientation), which were consolidated into a “combined_reward” metric. Exploratory data analysis revealed a notable performance shift around iteration 400, pointing to environmental or policy changes. Random forest regression yielded an R² of 0.9985, while logistic regression attained 100% accuracy in distinguishing high- and low-reward episodes at a -2 threshold. However, such near-perfect performance raises overfitting concerns, indicating the need for additional data diversity or real-world testing to validate generalizability. From a practical standpoint, these findings can reduce the cost and risk of physical experimentation, as they guide calibration of movement rates, sensor alignments, and reward function tuning before real-world deployment. By combining rigorous data analysis with a high-fidelity robotics simulator, this study shows how relatively simple algorithms can deliver powerful insights into robot-environment interactions, potentially saving significant development time. Overall, the project offers a simulation-first approach that lays the groundwork for further studies on adaptive grasp strategies, domain randomization, and scalable integration of data science in industrial automation.