Stock Investment Modeling and Prediction Using Vector Autoregression (VAR) and Cross Industry Standard Process for Data Mining (CRISP-DM)

Abstract

Prediction investment share is one issue important right now, as many people now have to switch to digital investment. Many studies have been done to help Stock data prediction using machine learning. However, most machine learning models used complex and predictive only one variable row time. This research focuses on creating machine learning models using the VAR algorithm to predict several variables at a time with 1 model and provides recommendations, and uses the framework Cross Industry Standard Process for Data Mining (CRISP-DM) work in holding his research. The contribution of this research is to analyze whether the open, high, low, and close share price variables can be predicted based on each variable’s past data. Then build a forecasting model using machine learning technology, the Vector Autoregression (VAR) algorithm, and the Cross Industry Standard Process for Data Mining (CRISP-DM) method. From the resulting study, the VAR model is able to produce a model capable of predicting three variables at a time, that is, price highs, lows, and closes with each R2 Score is 0.60, 0.51, 0.54 and uses an optimal lag of 273 but for the variable price opening make a separate model with the lag difference is two lags and the R2 Score is 0.63. Based on the results of testing and evaluation of the use of R2, MAE, and RMSE scores on the model that was successfully created, it can be concluded that VAR can be used to predict the highest, lowest, and closing stock prices at once and has a fairly good accuracy even though the opening price variable must make a separation.