Human Activity Recognition (HAR) is associated with quite significant applications in fitness tracking, disease prediction, and many more health related applications. Our study uses various deep learning techniques like Self-Organizing Maps (SOMs), Long Short-Term Memory (LSTM), Multilayer Perceptron (MLP), Convolutional Neural Networks (CNNs), and Deep Belief Networks (DBNs) to present a comprehensive comparison of the resultant results after the application of these on the HAR dataset. The relevant features were extracted from the dataset and then normalized for consistency of inputs across the different models. The results that we got indicated a varying degree of effectiveness depending on the technique employed. CNN and LSTM capture the environment well. While MLP provides feature extraction, CNN shows the power of spatial feature extraction. DBN can learn hierarchical representations better. This study provides a comparison of various deep learning methods on mobile phone sensor-based HAR datasets, providing insight for future researchers so that they can choose the most appropriate one according to their data usage. This information includes exercise and sports, health, safety, smart environment, transportation travel, etc., may form the basis for future research in these areas.