Comparative Analysis of Custom Neural Network Model and VGG-16 for Multi-class Image Classification

This study assesses the performance of CustomNet, a lightweight neural network model trained using NumPy and Pandas, compared to the VGG-16 architecture on the datasets of MNIST, Fashion MNIST, and Virus MNIST. With fewer parameters, CustomNet achieves competitive accuracy since it is designed for reduced computational complexity. Model performance was assessed using metrics such F1 score, accuracy, precision, and recall. CustomNet provides a compelling trade-off between computational efficiency and accuracy, making it appropriate for resource-constrained contexts, even if the VGG-16 model showed better accuracy.