We propose a robust encoder-decoder model based on Recurrent Neural Network (RNN) for the data-driven option hedging problems. This approach naturally incorporates sequential information and features selection. Using S&P 500 index option market data from January 2, 2004, to August 31, 2015, we demonstrate that the daily hedging performance of the proposed model surpasses that of the minimum variance quadratic hedging formula, corrective methods based on LVF and SABR models, and the previous data-driven kernel model. In addition, we demonstrate that the weekly and monthly hedging performance of the proposed model significantly surpasses that of the previous data-driven kernel model and Black-Scholes model with implied volatility.