The stock price received from KOSCOM (Korea Securities Computing Corporation), which provides Korean financial IT solutions, consists of forty items (four groups: domestic and foreign or buying and selling) such as domestic selling high price, foreign selling opening price, and domestic buying trading amount. For example, even if there are stock prices with the same value, their inside combination is different. Domestic selling high price is downturn and domestic buying trading amount is upturn whereas domestic selling high price is upturn and domestic buying trading amount is downturn. Because of very changeable items, the goal is to predict the next stock price pattern graph using these items and for this prediction to be of value. Analysis and prediction in the stock market are being studied using various methods, such as machine learning and text mining. First, regarding data mining studies using daily stock data, there are prediction research works based on support vector machine (SVM) ([8, 24]) in order to determine whether the new pattern data belongs to a certain pattern category, artificial neural network (ANN) ([31, 32]), to have good prediction even with a complex relationship between the variables, and autoregressive integrated moving average (ARIMA) ([40, 46]) to identify and predict time series variation. Unlike machine learning, there are several prediction research works based on the word analysis of news articles ([29, 38, 39]). As these research works have predicted daily stock prices using daily closing price, it is not sufficient to make predictions in a time period as short as an hour and a half. Moreover, even if they have analyzed the significance of variables and increased the prediction accuracy of the model by eliminating unimportant variables, the error rates of the prediction are higher because of the use of any The stock price graph consists of several patterns such as consolidation, cup with handle, double bottom, and saucer, as shown in Figure 1 [7, 49]. As these patterns repeatedly appear at fixed time intervals, finding a pattern parallel to the current pattern will enable prediction of the following pattern. Focusing on this point, in this paper, we propose a new method for generating stock price predictions based on historical stock big data. First, unlike existing studies that mostly use closing price data, we use tick-by-tick data for short term prediction and aggregate them to transform non-continuous data to continuous data. Then, through a dynamic time warping algorithm we make some patterns similar to the current pattern and select important features affecting the stock price by using stepwise regression with them.