The stock market is one of the extensively researched empirical topics in the financial sector. Many investors and researchers have used statistical and mathematical methods to conduct various studies to achieve market success, including stock price or earning predictions and optimal portfolio construction. With the development of deep learning technology, the learning and utilization of stock index has been actively researched. The stock index is a time-series data that includes noise and basic information. These noises work as significant obstacles to stock index learning. Therefore, the removal of this noise will result in more accurate stock index predictions. A typical denoising method is the moving average line method. Recently, the use of a denoising filter, used in signal processing, to remove noise in a stock index has been researched. These methods highlight the learning improvement in deep learning. However, the question of how to determine the parameters remains. In this paper, we present a time series denoiser based on long short-term memory (LSTM) stacked autoencoder and use it to predict the Korea Composite Stock Price Index (KOSPI) 200 stock index.