This thesis explores the quantification of social behavior for neural data analysis in mice. We developed an unsupervised machine learning pipeline to classify complex social behaviors from video recordings of resident-intruder experiments. The pipeline includes pose estimation, feature extraction, and dimensionality reduction techniques to create a low-dimensional behavioral embedding. We then integrated fiber photometry data from the medial amygdala (MeA) and bed nucleus of the stria terminalis (BNST) to investigate the neural correlates of social behaviors. Our analysis revealed sex-specific behavioral patterns and neural responses in these brain regions during social interactions. Using causal inference techniques, we found that both MeA and BNST neural activity showed stronger predictive power for future behavior than vice versa, suggesting their role in behavior generation rather than purely sensory processing.