The recognition of human activities through computational models is a pivotal field within machine learning, focusing on discerning various actions taken by people using data from sensors. This study report offers an in-depth evaluation of such recognition by integrating machine and deep learning strategies. The research work utilizes the universally accessible UCI repository's Human Activity Recognition using Smartphones dataset, alongside the more recent HAR70+ dataset, which is specifically designed for participants aged between 70 and 95 years. We employed a two-fold methodological framework for HAR: firstly, utilizing conventional machine learning models, and secondly, applying a deep learning strategy with a network based on convolutional Long Short-Term Memory (LSTM). The most effective machine learning model was the SVM, reaching accuracies of 95.046%, and the LSTM network achieved 98.097%. The findings indicate machine and deep learning methods are adept at distinguishing human actions via sensor data gathered from smartphones. The basic machine learning method attains high accuracy with reduced computational demand, whereas the deep learning method excels at processing intricate and nonlinear data structures. Our work provides a useful framework for future studies on HAR using sensor data from a smartphone, which can have significant applications in healthcare, sports, and security.