Journal international d'intelligence d'essaim et de calcul évolutif

Abstrait

Performance Analysis of Backpropagation Artificial Neural Networks with Various Activation Functions and Network Sizes

Hamed Hosseinzadeh*

This paper conducts a comprehensive performance analysis of Back Propagation Artificial Neural Networks (BP- ANNs) utilizing various activation functions. Activation functions play a crucial role in shaping neural networks’ behavior and learning capabilities. Through systematic evaluation across diverse network sizes (numbers of hidden layers and neurons), this study assesses the impact of commonly employed activation functions-such as Sigmoidal, Tanh, Clog log, Aranda, and others-on the convergence speed and accuracy of BP-ANNs. The findings provide empirical insights essential for optimizing neural network artificial intelligence architectures tailored to specific applications and datasets.