This study explores the use of machine learning (ML) models, specifically Support Vector Regression (SVR) and Gaussian Process Regression (GPR), to predict nuclear mass excesses using experimental data and a physics-based feature space. The research aims to assess the performance of these models in predicting nuclear mass excesses, both within the training and test regions and in extrapolation regions beyond the available data. The models are trained using experimental data from the Atomic Mass Evaluation 2020 (AME2020) and a physics-based feature space that includes 12 inputs such as Z (proton number), N (neutron number), A (mass number), and other nuclear properties. The results show that both SVR and GPR models perform well in predicting nuclear mass excesses, with GPR demonstrating better performance, especially when using a larger feature space. The models also show reasonable predictions outside the training and test regions, comparable to existing theoretical models. The study further evaluates the extrapolation capabilities of the models, showing that they can make accurate predictions in regions where experimental data is limited. Additionally, the study incorporates Explainable AI (XAI) techniques, such as SHapley Additive exPlanations (SHAP), to interpret the models' predictions and understand the contributions of different features. The results indicate that the GPR model is more effective in capturing the physics of nuclear mass excesses and provides more accurate predictions, particularly for medium-heavy and heavy nuclei. The study concludes that SVR and GPR can be reliable and efficient tools for predicting nuclear mass excesses and exploring nuclear properties in the future.This study explores the use of machine learning (ML) models, specifically Support Vector Regression (SVR) and Gaussian Process Regression (GPR), to predict nuclear mass excesses using experimental data and a physics-based feature space. The research aims to assess the performance of these models in predicting nuclear mass excesses, both within the training and test regions and in extrapolation regions beyond the available data. The models are trained using experimental data from the Atomic Mass Evaluation 2020 (AME2020) and a physics-based feature space that includes 12 inputs such as Z (proton number), N (neutron number), A (mass number), and other nuclear properties. The results show that both SVR and GPR models perform well in predicting nuclear mass excesses, with GPR demonstrating better performance, especially when using a larger feature space. The models also show reasonable predictions outside the training and test regions, comparable to existing theoretical models. The study further evaluates the extrapolation capabilities of the models, showing that they can make accurate predictions in regions where experimental data is limited. Additionally, the study incorporates Explainable AI (XAI) techniques, such as SHapley Additive exPlanations (SHAP), to interpret the models' predictions and understand the contributions of different features. The results indicate that the GPR model is more effective in capturing the physics of nuclear mass excesses and provides more accurate predictions, particularly for medium-heavy and heavy nuclei. The study concludes that SVR and GPR can be reliable and efficient tools for predicting nuclear mass excesses and exploring nuclear properties in the future.