In this work, we suggest a new approach to identify Micro Expressions (ME) in the facial region using Machine Learning, categorizing them into simulated and neutralized types. The study explores diverse techniques for feature extraction and pre-processing to emphasize unique patterns, aiming to improve model convergence, generalization, and mitigate overfitting. The proposed method recognizes challenges related to spontaneous subtle motion, environmental variations, and imbalanced datasets in Micro Expression Recognition. To tackle these issues, we incorporated the XGBoost classifier and a hybrid GA-PSO algorithm for optimizing feature selection, enhancing the model’s ability to differentiate and classify various classes within a dataset. The research utilizes the CASME II dataset, emphasizing its relevance in addressing real-world scenarios and validating the proposed approach. The combination of XGBoost and GA-PSO presents a promising direction for future advancements in this field.