Accurate weather classification is essential in numerous real-world applications, significantly impacting areas such as solar energy systems, outdoor events, and visual systems. For minimizing the adverse effects stemming from untimely weather fluctuations, its prediction is of utmost importance. The present work reflects an approach for multi-class weather classification by harnessing the power of deep learning techniques and transfer learning. The research utilizes the Multi-class Weather Dataset (MWD), comprising four weather conditions: sunrise, shine, rain, and cloudy. Using pre-trained deep learning architectures such as ResNet50, Densenet-161, InceptionV3, VGG19, and MobileNet, the present research extracts intricate features from the dataset images. Also, Transfer learning is employed to fine-tune these models for weather classification, effectively addressing the constraints posed by a scarcity of labeled data. The role of a logistic regression model is pivotal as the multi-class classifier. In-depth evaluation metrics encompassing accuracy, precision, recall, and F1 score are employed, accompanied by a meticulous analysis utilizing confusion matrix techniques. The results show that Densenet-161 demonstrates superior performance while other models exhibit competitive results presenting viable alternatives for specific use cases. Overall, the study contributes insights into the effectiveness of pre-trained models and transfer learning in multi-class weather classification tasks.