Can AutoML outperform humans? An evaluation on popular OpenML datasets using AutoML Benchmark