Recognizing social cues and emotions is vital for navigating daily interactions, understanding emotions in conversations, interpreting body language in meetings, and supporting friends in difficult situations. This work focuses on analyzing group-level emotions in videos captured in natural settings, marking an attempt at multimodal group-level emotion analysis. Automatic group emotion recognition is pivotal for understanding complex human-human interactions. Group emotion recognition in videos presents several challenges because existing work predominantly focuses either on individual emotion recognition or group emotion analysis in static images. To address this challenge, we introduce a deep-learning-based multimodal fusion model that integrates diverse modalities, including audio, video, and scene. Feature extraction employs advanced models like TimeSformer for video description and wav2vec2.0 for audio analysis. All the experiments are conducted on the VGAF dataset. Our key findings include: (1) Multimodal approaches outperform their unimodal counterparts, (2) Experimental results confirm the superior performance of proposed approach compared to benchmark methods on the given dataset, and (3) There is a strong correlation between modalities and respective emotions.