Classification of Traffic Signs using Neural Network Algorithms and Comparison of Algorithms

Abstract

The neural network algorithms are used to train different models for forecasting, classification, interpret, and analyze the results. The motive of the study was to train the data set with different neural network algorithms and check the classification accuracy and make comparison, which algorithm is best among all of them. The "German traffic sign" dataset is used to classify the different traffic signs. Each algorithm run for thirty times with different number of layers and hidden layers. A neural network depends on the learning rate, hidden layers, layers and activation function. The network gave calibration values, validation values, different results for different number of layers, and hidden layers. Smallest absolute gradient gave the best results for calibration (87.73) and validation (84.05) and at 2, 3,and 4 number of layers it gave the best classification accuracy. Smallest learning rate gave the least results for calibration (83.34) and validation (78.23). The outcome suggested that although the differences among the algorithms are not big, the SAG gave the highest classification accuracy.