This project presents a multi-stage computational framework to stream- line the analysis of rat behavior in conditioning experiments, a common pro- cedure in neuroscience and behavioral research. Traditional manual analysis of video-recorded sessions, which document rats responses to conditioned stimuli, is labor-intensive and prone to error. Our approach leverages deep learning to automate this process, enhancing both efficiency and accuracy in behavioral assessments. In the first stage, we use deep learning-based methods to seg- ment key rat body parts and detect the rearing posture across video frames. To train these models, we developed a novel semantic segmentation dataset, enabling the use of CNN-based architectures with supervised learning. Next, our method extracts spatio-temporal descriptors from the segmented frames, allowing for precise quantification of behavior over time. In the final stage, we generate visual representations of these descriptors, creating a comprehensive view of behavior patterns such as freezing, rearing, and grooming. This method not only reduces the manual workload but also provides a robust, data-driven approach to understanding complex behavioral responses in animal models, opening avenues for more consistent, large-scale behavioral research.