Empirical Analysis of Image Recognition and Classification through Machine Learning and Transfer Learning Approaches using Multiple Regression Analysis

Image recognition and classification have taken a shift from the conventional detection process by humans to detection by Machine Learning (ML) and Transfer Learning (TL). These two approaches are being applied in various image identification processes for human development. The research paper identified the accuracy rates of three architectures in Convolutional Neural Networks (CNNs); such as AlexNet, GoogleNet and ResNet. Different sources have been evaluated to collect the accuracy rates of these three CNN architectures. These architectures were trained with a definite number of resources having the same features (1000) and accuracy was calculated after every training day. The accuracy rate was kept on Excel and regression analysis was accomplished in SPSS to identify how one-day training improves the accuracy of these architectures. Moreover, training days and layers were identified for each architecture to compare the research data and to evaluate further. Findings suggested that AlexNet is more accurate in machine learning image recognition and classification, and it requires less time for training than GoogleNet and ResNet; whereas, due to having few layers, AlexNet is less accurate in Transfer Learning.