<p>Human activity recognition in crowded scenes remains a challenging task, especially under varying weather conditions. The key challenges are primarily related to image capture, content, and processing limitations. Varied weather conditions can limit the capture of images, resulting in human subjects appearing in videos of varying sizes, intentions, and occlusions, which make action identification more challenging. This paper presents a novel approach that combines the strengths of YOLOv5 and the extreme gradient-boosting classifier (XGBM) to accurately recognize group activities in diverse environments. The proposed model leverages the real-time object detection capabilities of YOLOv5 to extract spatiotemporal features from video sequences, which are subsequently fed into an XGBM classifier for robust activity recognition. The model evaluates on a comprehensive dataset comprising mall, crowd, and video sequences captured under different weather conditions. Experimental results with an ablation study demonstrate a significant improvement in accuracy and efficiency compared to state-of-the-art methods. Specifically, the proposed model achieves a mean average precision (mAP) of 92.1%, a precision of 90%, and a recall of 82.2% on the combined dataset, surpassing previous approaches of YOLOv5 by 2.2%, and also exhibits a 30 FPS frame rate, making it suitable for real-time applications. The advancement has the potential to enhance surveillance systems, autonomous vehicles, and other applications that rely on accurate human behavior understanding.</p>