1: \begin{abstract} \small\baselineskip=9pt 

2: In this paper, we propose the \texttt{StepDIRECT} algorithm for derivative-free optimization (DFO), in which the black-box objective function has a stepwise landscape. Our framework is based on the well-known \texttt{DIRECT} algorithm. By incorporating the local variability to explore the flatness, we provide a new criterion to select the potentially optimal hyper-rectangles. In addition, we introduce a stochastic local search algorithm performing on potentially optimal hyper-rectangles to improve the solution quality and convergence speed. Global convergence of the \texttt{StepDIRECT} algorithm is provided. Numerical experiments on optimization for random forest models and hyper-parameter tuning are presented to support the efficacy of our algorithm. The proposed \texttt{StepDIRECT} algorithm shows competitive performance results compared with other state-of-the-art baseline DFO methods including the original \texttt{DIRECT} algorithm. 

3: \end{abstract}

4: