Identifying diagnostic indicators for type 2 diabetes mellitus from physical examination using interpretable machine learning approach