Exploring New Horizons: Automated Machine Learning for Image Classification Networks

As machine learning technologies advance, simple image classification networks have become commonplace for developers like us in the field of neural networks. However, for humanities and social science researchers who are unfamiliar with artificial intelligence technologies, image classification remains a labor-intensive and error-prone task. The emergence of Automated Machine Learning (AutoML) technology has removed the barriers to deploying high-performance machine learning models for artists and art gallery managers who lack expertise in machine learning. For the traditional problem of image classifying, AutoML provides a simple and practical approach for building the most appropriate model based on relevant data in just one pass. In this study, we constructed a dataset for the classification of Traditional Chinese paintings and Western paintings. We then use the Tree-based Pipeline Optimization Tool for Automating Machine Learning (TPOT) algorithm to generate 1000 networks and select the model with the highest accuracy for the dataset. Our model attained a 99.29% accuracy rate on the dataset encompassing Traditional Chinese and Western paintings, demonstrating superior performance compared to commonly employed baseline models. We provide a standard AutoML practice paradigm, which simplifies the construction process of a complex machine learning model into four steps: data collection, feature engineering, model building, and model evaluation. This allows humanities researchers to deploy high-performance models without requiring profound knowledge in machine learning.