In recent years, artificial intelligence (AI), machine learning (ML), and deep learning have become crucial in image classification tasks. Traffic sign detection and classification, a key component of advanced driver assistance systems (ADAS), significantly enhance driving comfort and improve traffic conditions globally. This study focuses on implementing and evaluating deep learning models for the classification and recognition of Malaysian traffic signs. While convolutional neural networks (CNNs) are commonly used for image recognition, they typically require large datasets to achieve high accuracy. To address the challenge of limited data for Malaysian traffic signs, this study employs transfer learning—a technique that leverages pre-existing knowledge for new tasks. The study is structured into three phases: Phase 1 involves importing a pre-trained model; Phase 2 entails training the model using transfer learning techniques; and Phase 3 assesses the model's performance with both static and real-time images. The results demonstrate that the trained model achieved a training accuracy of 97.5% and a perfect prediction rate on test images. Key parameters for the final model included an initial learning rate of 0.001, a batch size of 32, and the Adam optimizer. The DenseNet-201 architecture was selected as the final model due to its superior performance and acceptable training time compared to other models.