Different kinds of study have been conducted on video footage utilizing deep learning approaches in Artificial Intelligence (AI). Behavior analysis, scene understanding, scene classification, object localization, event detection, and human activity recognition (HAR) make up the bulk of these tasks. Human Activity Recognition, or HAR, is one of the most challenging and consequential areas of study in the processing of video data. HAR may be applied in several domains, including video surveillance systems, human-computer interfaces, human behavior characterization, and robots. To enhance this approach and automatically identify pertinent characteristics, deep learning techniques have been employed. This study presents a detailed analysis of a machine learning model’s performance over ten epochs, utilizing training and validation datasets to assess accuracy and loss. The model exhibits progressive improvements in training accuracy, starting from approximately 55% and culminating at 71.7%, while validation accuracy follows a similar upward trend but remains consistently lower, indicating potential early signs of overfitting. This comprehensive analysis reveals that while the model is effectively learning and adapting to the dataset, The findings underscore the importance of continuous monitoring and adaptive adjustments in the training process to achieve a well-balanced model that accurately and consistently predicts on new data.