Comparative Analysis of Different Machine Learning Techniques on Traffic Signs Dataset

Abstract The rapid advancements in intelligent transportation systems have highlighted the critical importance of accurate and efficient traffic sign recognition for road safety. Traditional methods relying on handcrafted features and classical machine learning techniques have shown limited accuracy and robustness under varying environmental conditions. This study explores the potential of deep learning, specifically Convolutional Neural Networks (CNNs), for traffic sign recognition. We conduct a comparative analysis of various CNN architectures, including custom models and well-established networks like VGG16, VGG19, EfficientNetB0, and ResNet50. Additionally, we investigate hybrid models that combine features from different architectures to enhance accuracy and robustness. Using the GTSRB dataset, we evaluate the performance of these models based on metrics such as test accuracy, test score, and F1-score. Our findings indicate that custom CNN models and certain hybrid combinations, such as Custom CNN x VGG16, achieve superior performance. This research contributes to the academic understanding of machine learning techniques in traffic sign recognition and holds practical significance for improving the safety and efficiency of modern transportation networks.