Negatively charged nitrogen-vacancy (NV) centers in diamonds, as all-optical sensing, enable magnetic field sensing without microwave excitation, simplifying sensor design. This study explores an all-optical method using fluorescence lifetime measurements and evaluates regression models to infer magnetic field strength. A dataset of 22,758 observations is used to compare machine learning approaches, including symbolic regression, ensemble methods, and linear models. Results show that LightGBM, XGBoost, and Random Forest achieve near-perfect accuracy ( \(R^2 \approx 0.999\) ), while symbolic regression underperforms. Computational efficiency analysis highlights models like XGBoost and LightGBM provide an optimal balance between accuracy and execution time, making them strong candidates for deployment in resource-constrained environments. These findings advance all-optical magnetometry through data-driven approaches.