A Comprehensive Analysis of CNN-Based Deep Learning Models: Evaluating the Impact of Transfer Learning on Model Accuracy

This paper conducts a comprehensive analysis of Convolutional Neural Network (CNN)-based deep learning models, focusing on the impact of transfer learning on their accuracy. Multiple tailored CNN architectures are implemented for image classification tasks, serving as baseline models with carefully selected architecture and hyperparameters. To boost performance, transfer learning is introduced using pre-trained CNN models (ResNet, EfficientNet) on large-scale image datasets. Transfer learning is applied at various stages, and its influence on convergence speed and final accuracy is investigated. The study meticulously records and compares results between CNN models with and without transfer learning, exploring metrics like convergence, training time, and final accuracy. This study investigates the role of transfer learning in improving the performance of CNN-based image classification. The results provide actionable insights for computer vision researchers and practitioners, helping to optimize deep learning techniques for image classification and similar tasks. By understanding how transfer learning influences CNN models, this research contributes to advancing efficiency and innovation in the field.