Human Action Recognition (HAR) has emerged as a challenging research domain garnering significant attention within the computer vision community. This paper introduces a novel framework for action recognition utilizing Semi-supervised Generative Adversarial Networks (GANs). HAR involves identifying human actions depicted in videos, presenting inherent difficulties due to diverse visual and motion characteristics, changes in camera viewpoints, dynamic backgrounds, noise, and large datasets. We commence by scrutinizing and comparing prevalent state-of-the-art techniques before proposing our approach. The conventional stages in action recognition encompass object and human segmentation, feature extraction, activity detection, and classification. Our method leverages Generative Adversarial Networks, specifically employing a novel Semi-supervised GAN architecture fine-tuned with Support Vector Machine (SVM). Experimental results demonstrate the superior performance of our proposed technique over existing state-of-the-art methods, particularly validated on the UCF101 dataset, a widely-used benchmark for challenging human action recognition tasks.