There have been substantial advances in HAR (Human Activity Recognition) in the past several years due to the advancement of the IoT (Internet of Things). HAR may be used in a range of contexts, including elder care, surveillance systems, and anomalous behavior detection. Different machine learning techniques have been used to anticipate human actions in a particular circumstance. Feature engineering techniques, which may pick an ideal collection of features, have outperformed typical machine learning techniques. Deep learning models, like CNN (Convolutional Neural Networks), on the other hand, are known to extract features and minimize computing costs automatically. We employ the CNN model for predicting actions from the Weizmann Dataset in this article. To extract deep image features and trained machine learning classifiers, transfer learning is used in particular. We found that VGG-16 has an accuracy of 96.95 percent in our experiments. We also found that VGG-16 outperformed the rest of the CNN models that were used in our experiments.