This research investigates the efficacy of various machine and deep learning models in the domain of Human Activity Recognition (HAR), utilizing datasets from diverse sources including GBA, IXMAS, WVU, KTH, and WEIZMANN. Through meticulous analysis, traditional models like Decision Trees and GaussianNB demonstrated notable capabilities in recognizing human activities, albeit with limitations in handling complex temporal relationships inherent in activity data. This identified gap underscores the necessity for models that can intricately understand temporal dynamics and offer higher precision in activity classification. In response, we propose the utilization of Bi-Directional Long Short-Term Memory (Bi-LSTM) networks, a deep learning approach known for its proficiency in capturing long-term dependencies within sequential data. The results were compelling, as the Proposed Bi-LSTM model consistently surpassed traditional and other advanced machine learning models across all evaluated metrics—Accuracy, Precision, Recall, and F1 Score—across every dataset. Notably, on the KTH dataset, the proposed Bi-LSTM model achieved an unprecedented accuracy of 99.56% and an F1 score of 99.45%, illustrating a significant advancement over existing methods.